System and method for controlling a production plant consisting of several plant components, in particular a production plant for producing industrial goods such as semi-finished metal products

Description

[0001] The invention relates to a system and method for controlling a production plant consisting of several plant components, in particular a production plant for the production of industrial goods such as semi-finished metal products. The invention particularly relates to production plants for the manufacture of metal slabs, sheets or strips, beams or pipes, as well as their precursors, or for carrying out individual production steps in which product quality must be ensured.

[0002] Traditionally, product manufacturing takes place as a sequence of individual processes. Each of these processes is governed by one or more technological parameters, compliance with which is intended to ensure the desired product properties are achieved. Compliance with these technological parameters is ensured by the various control and regulation systems, generally by the plant's automation system.

[0003] The technological target specifications are usually determined through a sequence of theoretical considerations, simulations, laboratory tests, and finally, individual operational trials. Such recipes are usually static and are not adjusted during the production sequence, or only in exceptional cases, through manual intervention.

[0004] Due to non-compliance with or poorly selected technological specifications, defects in products occur within industrial production, which mean that they cannot be used for the purposes for which they were originally intended, or in the worst case, cannot even be put to commercial use at all.

[0005] An evaluation of production parameters during manufacturing and / or achieved qualities typically takes place manually and a link between an evaluation and technological target specifications is not automated.

[0006] To improve the manufacturing of products in a production plant with multiple plant components, US Patent No. 10,365,640 B2 proposes using machine learning to determine the relationships between the individual plant components and to derive control rules based on recorded quality parameters that intervene in the plant control system. The parameters of the individual plant components are recorded using sensors, and the control rules are implemented via actuators in the plant components. This results in cross-plant monitoring with control interventions in the plant control system. However, the plant control system is not generally optimized; rather, it is only intervened in by adapting control rules.

[0007] DE 199 30 173 A1 discloses a method for setting process parameters for a given product quality.

[0008] Based on this prior art, the present invention is based on the object of optimizing production in a production plant consisting of several plant components across all plants.

[0009] The object is achieved according to the invention by a system according to claim 1.

[0010] The production plant, in particular the multiple plant components of the production plant, each have separate plant automation systems for monitoring and controlling and / or regulating the production process within the production plant, in particular for monitoring and controlling and / or regulating the process within the plant components. The plant automation systems of the multiple plant components are independent of one another, and no information is exchanged between the plant automation systems of the multiple plant components.

[0011] A production plant control system generates target specifications for the automation of multiple plant components. In the current state of the art, production plant control is carried out separately for each component of the production plant.

[0012] Furthermore, a production facility typically has a production planning system with information about the products to be manufactured in the facility, particularly with target criteria for the products to be manufactured. The production planning system preferably operates across all production facilities, i.e., for all plant components. The production planning system contains information about the total products to be manufactured in the production facility, i.e., by all plant components.

[0013] The system according to the invention further comprises a model generator for generating at least one prediction model for the product currently being manufactured in the production plant and / or for the products to be manufactured in the future according to the production planning system. The prediction model relates in particular to the prediction of the properties of the products currently or in the future that are essential for the target criteria. The model generator takes into account the results of the monitoring of the production plant, in particular of the multiple plant components, when generating the at least one prediction model. The prediction model generated by the model generator thus provides a prediction regarding the current and future production process in the production plant, in particular a prediction of the product properties relevant to the target criteria, e.g. based on the type of manufacturing process, represented by process parameters characteristic of this process.The prediction model covers the entire production plant, taking into account the multiple plant components.

[0014] The model generator can, for example, use sensor and / or actuator data from the plant automation system, especially from multiple plant components. Using sensor data from the plant automation system improves the accuracy of the generated prediction model, while technological target specifications for the actuators or the individual control or regulation systems are available at an early stage of production.

[0015] The at least one prediction model determines a probability that one or more target criteria lie within a defined range. The prediction model thus determines a probability that one or more target criteria of the product currently manufactured in the production facility and / or products to be manufactured in the future lie within a defined range. Based on the determined probability, it can be determined whether the manufactured product is suitable for the intended use, i.e. whether it has the specified product quality. Alternatively, the prediction model can also predict a single value for a property relevant to the target criteria. In this case, the check is carried out, for example, by determining the distance between a property relevant to a target criterion and the respective target value or by monitoring whether a lower or upper limit for this property is exceeded.

[0016] The system according to the invention further comprises a production optimizer for determining an optimized production process within the production plant based on the data from the plant automation system, the production planning system, and the predictive model generated by the model generator. The production optimizer takes into account the production-technical specifications of the individual plant components when determining the optimized production process within the production plant. Thus, a cross-production plant optimization of the production process takes place, which takes into account all information from the other systems of the production plant. In comparison to the control of a production process known from US 10,365,640 B2, not only is the production process regulated, but a cross-production plant optimization takes place.

[0017] According to a variant of the invention, the production plant is a plant in the metal-producing industry or the steel industry.

[0018] According to a further variant of the invention, the plurality of plant components are selected from: electric arc furnace, blast furnace, converter, ladle furnace, vacuum ladle treatment, continuous casting plant, foundry, hot rolling mill, casting and rolling plant, reheating furnace, pickling line, cold rolling mill, annealing line, galvanizing line, tinning line, painting line, cross-cutting and slitting line, tube rolling mill, beam rolling mill, drop forging, open-die forging and / or straightening machine.

[0019] In a preferred variant of the invention, the multiple plant components each comprise sensors for detecting product properties, process parameters, and / or operating states within the plant component. Product properties, process parameters, and / or operating states can thus be detected separately in each plant component. The detection of product properties, process parameters, and / or operating states is carried out using suitable sensors. In particular, the detection of product properties, process parameters, and / or operating states occurs continuously during product manufacture. However, the measurement of certain relevant properties only takes place at the end of the production process in the laboratory, in particular on previously separated samples.

[0020] According to a practical variant of the invention, the multiple system components each comprise actuators for adapting the production process within the system component. The actuators implement the target specifications for the system automation. The target specifications are therefore control commands for the actuators of the system components.

[0021] According to a variant of the invention, the plant automation accesses the data of the sensors and / or actuators of the multiple plant components and / or transmits data to the sensors and / or actuators of the multiple plant components.

[0022] In an advantageous variant of the invention, the system further comprises a central data storage for the individual components of the system, wherein the individual components of the system can access the data and, in particular, adapt the data. The central data storage thus simplifies the exchange of data between the individual parts of the system according to the invention, in particular between the multiple system components of the production plant and the higher-level components of the system according to the invention. The central data storage can be located within the digital infrastructure of the production plant or externally as a so-called cloud data storage.

[0023] According to an expedient variant of the invention, the target criteria are selected from: thickness, width, length, weight, tensile strength, yield strength, modulus of elasticity, elongation at break, corrosion resistance, presence or number of surface defects of various types, number of cracks on the surface or within the material, DWTT results, Charpy results, transition temperatures from ductile to brittle fractures, layer thickness of the zinc layer, pipe wall thickness, eccentricity, web height, flange height, flange thickness, profile, flatness and the like.

[0024] According to an advantageous variant of the invention, the model generator is based on methods of statistics, machine learning, artificial intelligence or the like to generate the prediction model.

[0025] In a particularly advantageous variant of the invention, the model generator generates a plurality of mutually different prediction models, wherein the different prediction models differ, for example, with regard to the creation method, the original data and / or the learning algorithms. In particular, a plurality of prediction models are generated for the same target criterion, which, for example, take into account different original data (input parameters). It is expedient to use different prediction models with different sets of input parameters, particularly when different data about a product is available at different times during production, so that some of the prediction models are already applicable, while for other prediction models the necessary input parameters are not yet fully available because the process step in which they are generated has yet to be carried out.

[0026] According to a variant of the invention, the model generator can use different sets of input variables to create the multiple, different prediction models. These can be values ​​for which target specifications exist within the framework of automation (specified variables) or measured values ​​or signals that arise reactively from the production process (measured data). These values ​​are expediently summarized process-step by process so that prediction models are generated that are usable right from the start of production because they only use specified variables, or prediction models for which all the measured data required for the prediction model is available after a specific process step. This can be advantageous because reactive measured data sometimes contains information that is otherwise difficult to capture and thus enables a more precise prediction model.

[0027] For any set of input parameters, a variety of predictive models can generally be created, such as neural networks, decision trees, support vector machines, linear models, discriminant analysis, Bayesian estimators, nearest neighbors, nonlinear multivariate regression, splines.

[0028] Different evaluation metrics can be used for selection. From the list of created prediction models, a prediction model is selected for each step in which the target specifications for automation are to be updated, which the optimizer then uses. Evaluation metrics can include the mean squared error, the mean absolute error, the coefficient of determination, or similar.

[0029] According to an expedient variant of the invention, the process optimizer takes into account the prediction model which provides the currently best prediction for the product currently manufactured in the production plant and / or for the products to be manufactured in the future according to the production planning system, in particular taking into account one or more target criteria.

[0030] Before the production process begins, there is naturally no measurement data available for a specific product. The optimizer therefore uses a predictive model that only uses input parameters for which target values ​​also exist within the automation framework. During the actual optimization step (preferably taking boundary conditions into account), the target values ​​are determined in such a way that the deviation of one or more target variables is minimized or the probability that the target variable leaves a validity range is minimized. Other optimization goals of a similar nature are also feasible. The production process uses the target values ​​determined in this way within the automation framework, with the control and regulation systems attempting to implement the target values ​​as far as possible.

[0031] During the first process step, the automation system records a large amount of measured data. This includes both those variables for which target values ​​existed, as well as other measured data. Generally, a variety of effects lead to deviations between the actual values ​​and the specified values. Through this, and possibly through the use of a predictive model that also uses additional measured data as input variables, the optimizer determines the target values ​​for all subsequent process steps before the start of the next process step, with the aid of a possibly different predictive model.

[0032] This process can be continued up to the final process step. Overall, production is optimized according to the best current understanding of the influence of the target parameters on the quality criteria, taking into account all relevant quality criteria. This results in fewer devaluations or complaints and improves the cost-effectiveness of the production process.

[0033] After a product has been completed, a quality assessment usually takes place, including testing samples of the product for certain quality attributes, while other quality attributes are also measured directly during production. This may, but does not necessarily, apply to all manufactured products.

[0034] As soon as a new quality rating is available for a product, the model generator can begin generating new predictive models again, as new knowledge about the causal relationships in the newly acquired production data and the quality rating is available. Generally, it is sufficient to perform this step after a series of manufactured products, rather than after each product.

[0035] By regularly updating the forecast models, changing production conditions can be taken into account and it is ensured that optimization is always aligned with the current mechanisms of action.

[0036] In a preferred variant of the invention, the process optimizer performs an evaluation with respect to one or more target criteria when determining the optimized production process within the production plant. Optimization with respect to multiple target criteria is particularly preferred when there is a connection between the individual target criteria. It is often the case that optimization with respect to one target criterion simultaneously influences another target criterion. These interconnected target criteria should be considered together during optimization, since otherwise, the separate optimization of one target criterion could cause other related target criteria to exceed a permissible value range.

[0037] According to a suitable variant, the evaluation is based on one of the following evaluation functions: mean absolute error, mean square error, summed loss of value per deviation of the target criteria, probability of leaving the tolerance range, or the like.

[0038] According to a further expedient variant of the invention, the process optimizer is based on methods of linear or quadratic programs, genetic optimization, reinforcement learning with Q-tables, neural networks, simulated annealing, Metropolis, swarm algorithms, hill climbing, Lagrange multiplier method or the like.

[0039] In an advantageous variant of the invention, the process optimizer determines the effects of setpoint changes on the target specifications of the product currently manufactured in the production plant and / or the product to be manufactured in the future according to the production planning system. The process optimizer thus determines whether a change in the target specifications for the plant automation results in the at least one or more target criteria that form the basis for the optimization meeting the specified requirements.

[0040] According to a particularly advantageous variant of the invention, the process optimizer comprises information on possible setpoint changes of the production plant, in particular of the individual plant components. The process optimizer preferably has information from the production plant, in particular of the individual plant components, implementable setpoint changes such as maximum rates of change of actuator activity and / or dependencies of an adjustable variable on other, likewise adjustable variables. This enables the process optimizer to determine the setpoint changes that can be implemented by the production plant, in particular of the plant components. Based on the implementable setpoint changes, the target specifications for the plant automation can be created and forwarded to the plant automation. The change rates are relevant both for switching between different products and for changes in the input data and thus the predictions regarding the product volume.Common cases include fluctuations along the length or width of a strip-shaped product. In this case, the optimizer can also generate different sets of target values ​​so that the target criteria for the individual product segments are met. The restrictions due to the maximum rates of change take into account that the process cannot generally proceed very differently for two adjacent product segments. According to an advantageous variant of the invention, the process optimizer optimizes the production process within the production plant for a plurality of product sections. These can be sections across the length or width of a strip-shaped product, for example, or areas such as flanges or webs of beam-like products.

[0041] The object is further achieved by a method according to claim 11.

[0042] According to the method according to the invention, information about the products to be manufactured in the production plant is recorded during product production. This relates to target criteria for the products to be manufactured. The purpose of the method according to the invention is to manufacture the products while adhering to the target specifications and with optimal utilization of the production plant. The information about the products to be manufactured in the production plant is usually stored in a production planning system and is queried or received from there. The production planning system preferably works across all production plants, i.e. for all plant components jointly. The production planning system contains information about the total products to be manufactured in the production plant, i.e. by all plant components.

[0043] The production plant preferably consists of several independent but consecutive plant sections.

[0044] The production process in the production plant is monitored, particularly within the multiple plant components. Monitoring takes place, for example, within the plant automation system in the individual plant components. The plant automation system is designed to monitor and control and / or regulate the production process within the production plant, particularly to monitor and control and / or regulate the process within the plant components. The plant automation systems of the multiple plant components are independent of one another, and no information is exchanged between the plant automation systems of the multiple plant components.

[0045] According to the method according to the invention, at least one prediction model is created for the product currently manufactured in the production plant and / or for the products to be manufactured in the future. The prediction model relates in particular to compliance with the target criteria for the products currently or in the future. The acquired information from the production plant, in particular from the multiple plant components, is taken into account when generating the at least one prediction model. The prediction model generated by the model generator thus provides a prediction regarding the current and future production process in the production plant. The prediction model relates to the entire production plant, taking the multiple plant components into account.

[0046] The model generator can, for example, use sensor and / or actuator data from the plant automation system, especially from multiple plant components. Using sensor data from the plant automation system improves the accuracy of the generated prediction model, while technological target specifications for the actuators or the individual control or regulation systems are available at an early stage of production.

[0047] The at least one prediction model determines the probability that one or more target criteria lie within a defined range. The prediction model thus determines the probability that one or more target criteria of the product currently manufactured in the production facility and / or of products to be manufactured in the future lie within a defined range. Based on the determined probability, it can be determined whether the manufactured product is suitable for the intended use, i.e., whether it has the specified product quality.

[0048] Furthermore, according to the invention, an optimized production process is determined which regulates production within the production plant based on the recorded information about the products to be manufactured in the production plant, the monitoring of the production process within the production plant, in particular within the plant components, the at least one generated prediction model, and production-related parameters of the production plant, in particular production-related parameters of the individual plant components. Thus, a cross-production plant optimization of the production process takes place, which takes into account all information from the other systems of the production plant. In comparison to the control of a production process known from US 10,365,640 B2, not only is the production process intervened in a regulatory manner, but a cross-production plant optimization takes place.

[0049] Based on the optimized production process, target specifications for the control and / or regulation of the production process within the production facility are generated and implemented. This particularly applies to the control and / or regulation of the processes within the plant components of the production facility.

[0050] According to a variant of the invention, the production plant is a plant in the metal-producing industry or the steel industry.

[0051] According to a further variant of the invention, the plurality of plant components are selected from: electric arc furnace, blast furnace, converter, ladle furnace, vacuum ladle treatment, continuous casting plant, foundry, hot rolling mill, casting and rolling plant, reheating furnace, pickling line, cold rolling mill, annealing line, galvanizing line, tinning line, painting line, cross-cutting and slitting line, tube rolling mill, beam rolling mill, drop forging, open-die forging and / or straightening machine.

[0052] In a preferred variant of the invention, the production process within the production plant, in particular within the plant components, is monitored by means of sensors. Particularly preferably, the production process within the production plant, in particular within the plant components, is monitored continuously.

[0053] According to a suitable variant of the invention, the control and / or regulation of the production process within the production plant, in particular the control and / or regulation of the processes within the plant components, is carried out by means of actuators. This allows the generated target specifications to be implemented automatically.

[0054] According to an advantageous variant, the method according to the invention further comprises the step of storing data in a central data storage device, in particular data relating to the acquisition of information about the products to be manufactured in the production plant, in particular with target criteria for the products to be manufactured, data relating to the monitoring of the production process within the production plant, in particular within the plant components, data relating to the generation of at least one prediction model for the product currently manufactured in the production plant and / or for the products to be manufactured in the future, in particular with regard to compliance with the target criteria for the products currently or in the future to be manufactured, data relating to the determination of an optimized production process within the production plant on the basis of the acquired information about the products to be manufactured in the production plant,Monitoring the production process within the production plant, in particular within the plant components, the at least one generated prediction model and production-related parameters of the production plant, in particular production-related parameters of the individual plant components, and / or data relating to the generation and execution of target specifications for the control and / or regulation of the production process within the production plant, in particular the control and / or regulation of the processes within the plant components. Central data storage simplifies the exchange of data between the individual parts of the production plant, in particular between the multiple plant components of the production plant and the higher-level components of a system according to the invention designed to carry out the method according to the invention.

[0055] In a variant of the invention, the target criteria are selected from: thickness, width, length, weight, tensile strength, yield strength, modulus of elasticity, elongation at break, corrosion resistance, presence or number of surface defects of various types, number of cracks on the surface or within the material, DWTT results, Charpy results, transition temperatures from ductile to brittle fractures, layer thickness of the zinc layer, pipe wall thickness, eccentricity, web height, flange height, flange thickness, profile, flatness and the like.

[0056] According to an advantageous variant of the invention, the generation of the at least one prediction model is based on methods of statistics, machine learning, artificial intelligence, or the like. Such methods have the advantage that they continuously improve, particularly with increasing data volumes. The methods can be continuously trained, i.e., improved, thereby continuously improving the accuracy of the created prediction model.

[0057] According to a particularly advantageous variant of the method according to the invention, the method comprises the step of generating several different prediction models, wherein the different prediction models differ, for example, with regard to the generation method, the original data, and / or the learning algorithms. In particular, several prediction models are generated for the same target criterion, which, for example, take into account different original data (input parameters).It is useful to use different prediction models with different sets of input parameters, especially when different data about a product are available at different times during production, so that some of the prediction models are already applicable, while for other prediction models the necessary input parameters are not yet fully available because the process step in which they are generated has yet to be executed.

[0058] According to a variant of the invention, different sets of input variables can be used to create the multiple, different prediction models. These can be values ​​for which target specifications exist within the framework of automation (specified variables) or measured values ​​or signals that arise reactively from the production process (measured data). These values ​​are expediently summarized process-step by process, so that prediction models are generated that are usable right from the start of production because they only use specified variables, or prediction models for which all the measured data required for the prediction model is available after a specific process step. This can be advantageous because reactive measured data sometimes contains information that is otherwise difficult to capture and thus enables a more precise prediction model.

[0059] For each set of input parameters, a variety of prediction models can generally be created, such as neural networks, decision trees, support vector machines, linear models, discriminant analysis, Bayesian estimators, nearest neighbors, nonlinear multivariate regression, and splines. Different evaluation metrics can be used for selection. From the list of created prediction models, a prediction model is selected for each step in which the target specifications for automation are to be updated, which the optimizer then uses. Evaluation metrics can include the mean squared error, the mean absolute error, the coefficient of determination, or similar.

[0060] In a suitable variant of the invention, the prediction model that provides the currently best prediction for the product currently manufactured in the production facility and / or for the products to be manufactured in the future is taken into account when determining the optimized production process within the production facility. This is done, in particular, taking into account one or more target criteria.

[0061] Before the production process begins, there is naturally no measurement data available for a specific product. Optimization therefore uses a predictive model that only uses input parameters for which target values ​​also exist within the automation framework. During the actual optimization step (preferably taking boundary conditions into account), the target values ​​are determined in such a way that the deviation of one or more target variables is minimized or the probability that the target variable leaves a validity range is minimized. Other optimization goals of a similar nature are also feasible. The production process uses the target values ​​determined in this way within the automation framework, with the control and regulation systems attempting to implement the target values ​​as far as possible.

[0062] During the first process step, the automation system records a large amount of measured data. This includes both those variables for which target values ​​existed, as well as other measured data. Generally, a variety of effects lead to deviations between the actual values ​​and the specified values. Through this, and possibly through the use of a predictive model that also uses additional measured data as input variables, the optimizer determines the target values ​​for all subsequent process steps before the start of the next process step, with the aid of a possibly different predictive model.

[0063] This process can be continued up to the final process step. Overall, production is optimized according to the best current understanding of the influence of the target parameters on the quality criteria, taking into account all relevant quality criteria. This results in fewer devaluations or complaints and improves the cost-effectiveness of the production process.

[0064] After a product has been completed, a quality assessment usually takes place, including testing samples of the product for certain quality attributes, while other quality attributes are also measured directly during production. This may, but does not necessarily, apply to all manufactured products.

[0065] As soon as a new quality rating is available for a product, the model generator can begin generating new predictive models again, as new knowledge about the causal relationships in the newly acquired production data and the quality rating is available. Generally, it is sufficient to perform this step after a series of manufactured products, rather than after each product.

[0066] By regularly updating the forecast models, changing production conditions can be taken into account and it is ensured that optimization is always aligned with the current mechanisms of action.

[0067] According to a preferred variant of the invention, when determining the optimized production process within the production plant, an evaluation is carried out with respect to one or more target criteria. Optimization with respect to multiple target criteria is particularly preferred when there is a connection between the individual target criteria. It is often the case that optimization with respect to one target criterion simultaneously influences another target criterion. These interconnected target criteria should be considered jointly during optimization, since otherwise, the separate optimization of one target criterion could cause other related target criteria to exceed a permissible value range.According to an expedient variant of the invention, the evaluation is based on one of the following evaluation functions: mean absolute error, mean square error, summed loss of value per deviation of the target criteria, probability of leaving the tolerance range, or the like.

[0068] According to the invention, the at least one prediction model determines the probability that one or more target criteria lie within a defined range. This probability can be used to determine whether the product manufactured in the production facility meets the required quality requirements with sufficient probability, since the quality of the manufactured product correlates with compliance with specified target criteria.

[0069] According to an expedient variant of the invention, the determination of the optimized production process within the production plant is based on methods of linear or quadratic programs, genetic optimization, reinforcement learning with Q-tables, neural networks, simulated annealing, Metropolis, swarm algorithms, hill climbing, Lagrange multiplier method or the like.

[0070] According to an advantageous variant of the invention, determining the optimized production process within the production plant includes examining the effects of setpoint changes on the target specifications of the product currently manufactured in the production plant and / or the product to be manufactured in the future. When determining the optimized production process, it is therefore examined whether a change in the target specifications for the plant automation results in the at least one or more target criteria that form the basis for the optimization meeting the specified requirements.

[0071] In a preferred variant of the invention, the determination of the optimized production process within the production plant takes into account information on possible setpoint changes of the production plant, in particular of the individual plant components. This particularly concerns setpoint changes that can be implemented by the production plant, in particular of the individual plant components, such as maximum rates of change of actuator activity and / or dependencies of an adjustable variable on other, similarly adjustable variables.

[0072] According to an advantageous variant of the invention, the method according to the invention optimizes the production process within the production plant for a plurality of product sections. These can, for example, be sections across the length or width of a strip-shaped product, or areas such as flanges or webs of carrier-like products.

[0073] The invention particularly relates to a system for carrying out the method according to the invention. The system can be implemented at least partially by software executed by a computing device, wherein the computing device is designed to control a production plant.

[0074] The invention is explained in more detail below using an exemplary embodiment illustrated in the figure. It shows: Fig. 1 a block diagram of a system according to the invention for controlling a production plant.

[0075] Figure 1 shows a block diagram of a system 1 according to the invention for controlling a production plant 2. The production plant 2 consists of several plant parts 3. In particular, the invention relates to a control system 1 for a production plant 2 for producing industrial goods such as metallic semi-finished products, such as a plant in the metal-producing industry or the steel industry.

[0076] The plant components 2 of the production plant 2 are, for example, an electric arc furnace, a blast furnace, a converter, a ladle furnace, a vacuum ladle treatment plant, a continuous casting plant, a foundry, a hot rolling mill, a casting and rolling plant, a reheating furnace, a pickling line, a cold rolling mill, an annealing line, a galvanizing line, a tinning line, a painting line, a cross-cutting and slitting line, a tube rolling mill, a beam rolling mill, a drop forge, an open-die forge, and / or a leveling machine. Generally, a plant component 2 within the meaning of the invention is a spatially and / or functionally definable part of the production plant 2. The plant components 2 can in turn be subdivided into various units, such as a scale washer, roughing stand, shear, finishing stand, cooling line, and coiler.

[0077] Production facility 2 is used to manufacture products. Products are individually identifiable production units that move through production facility 2, in particular sequentially through the plant components 3 of production facility 2. The products are subject to various process steps. In principle, an n:m relationship (with integers n, m) can exist between input and output products of production facility 2 and / or plant components 3, for example, by splitting and linking products.

[0078] A process is a separately identifiable work step that leads to a change in the shape or the internal or external properties of a product.

[0079] The production plant 2, in particular the individual plant components 3, comprise a plant automation system 4. The plant automation system 4 serves to monitor and control and / or regulate the production process within the production plant 2 or the plant components 3. The plant automation system 4 comprises sensors 10 and actuators 11 for monitoring and control and / or regulation.

[0080] A sensor 10 within the meaning of the invention is a measuring device within the production plant 2 or the plant components 3, which provides information about the processed products or the operating status of the production plant 2 or the plant components 3. This can also be information from secondary processes, such as laboratory measurements on product properties or the plant status, so that information about product quality is also available. The fact that a manual transmission step may be required should not constitute a limitation in this context. In general, a sensor 10 is used to record product properties and / or operating states within the production plant 2 and / or the plant components 3.

[0081] An actuator 11 within the meaning of the invention is an adjusting device by means of which the operating state of the production plant 2 or the plant components 3 can be directly or indirectly influenced, which can simultaneously affect the product being processed. An actuator 11 is therefore used in particular to adjust the production process within the production plant 2 and / or the plant components 3.

[0082] The system according to the invention can further comprise a sensor list 12 and / or actuator list 13. The sensor list 12 contains, for example, identifiers for sensors 10 that are relevant for controlling production in the production plant 2 or the plant components 3. The sensor list 12 can comprise all or only some of the sensors 10 of the production plant 2 or the plant components 3. Accordingly, the actuator list contains, for example, identifiers for actuators 11 that are relevant for controlling production in the production plant 2 or the plant components 3. The actuator list 13 can comprise all or only some of the actuators 11 of the production plant 2 or the plant components 3.

[0083] Automation, within the meaning of the invention, refers to the entirety of all control and regulation processes required to operate the production plant 2 or its plant components 3, as well as the necessary hardware. Among other things, the automation uses sensors 10 and actuators 11.

[0084] Plant automation 4 generally requires specifications regarding how a work step is to be performed to produce a product. These specifications are referred to as setpoints.

[0085] The plant automation 4 accesses in particular data from the sensors 10 and / or actuators 11 or transmits data to them.

[0086] The system 1 according to the invention or the production plant 2 comprises a production planning system 5 with information about the products to be manufactured in the production plant 2. In particular, the production planning system 5 contains target criteria for the products to be manufactured.

[0087] A number of properties play a role in the manufacture of certain products and their marketability. According to the invention, all relevant properties are referred to as target criteria. The target criteria are selected, for example, from: thickness, width, length, weight, tensile strength, yield strength, modulus of elasticity, elongation at break, corrosion resistance, presence or number of surface defects of various types, number of cracks on the surface or within the material, DWTT results, Charpy results, transition temperatures from ductile to brittle fractures, layer thickness of the zinc layer, pipe wall thickness, eccentricity, web height, flange height, flange thickness, profile, flatness, and the like.

[0088] Failure to meet a target criterion may require a product to be reassigned to another use, which generally entails lower remuneration. In the worst case, the product cannot be used for any purpose and must be disposed of or recycled.

[0089] The system 1 according to the invention further comprises a model generator 6 for generating at least one prediction model 7 for the product currently manufactured in the production plant 2 and / or for the products to be manufactured in the future according to the production planning system 5. The generated prediction model 7 relates in particular to compliance with the target criteria for the products currently or in the future. When generating the at least one prediction model 7, the model generator 6 takes into account the results of the monitoring of the production plant 2, in particular of the multiple plant components 3.

[0090] The model generator 6 establishes a relationship between sensor and actuator data and the product quality of the manufactured product. The model generator 6 is based, for example, on methods of statistics, machine learning, artificial intelligence, or the like. This process can generally be understood as a function that assigns a one- or multi-dimensional vector to an n-dimensional vector. The multi-dimensional case refers to the fact that, if necessary, several target criteria should be described simultaneously, but does not represent a restriction. To generate at least one prediction model 7, relationships to individual sensory or actuator parameters are examined. This can be done by checking correlation, shared information, Shapley values, feature ranking, or similar, or simply model-intrinsic.Some of the parameters listed in the sensor and actuator lists 12, 13 can be sorted out as irrelevant for the respective target criterion, but this does not represent a limitation. In addition to the different model types, these can also differ in the choice of fundamental parameters ("hyperparameters" - e.g., the number and connection of neurons in neural networks). The quality of a prediction model 7 is defined by a metric; common variants include, among various others, the L1 or L2 measure ("mean absolute error" or "mean squared error"). It is important that the metric evaluates how well a prediction model 7 is able to predict the target criteria from the sensor and actuator data. The evaluation metric used is defined for each target criterion.When selecting parameters, particular emphasis can be placed on distinguishing between actuators 11 and sensors 10, since only the actuators 11 can influence the process and the product.

[0091] A trained prediction model 7 uses all or a selection of the sensor and actuator data to make a prediction for one or more of the target criteria. The prediction should at least include an expected value for the target criterion, but can also include a probability that the target criterion exceeds or falls below a certain critical minimum or maximum value.

[0092] The system 1 according to the invention further comprises a production optimizer 8 for determining an optimized production process within the production plant 2 based on the data from the plant automation system 4, the production planning system 5, and the prediction model 7 generated by the model generator 6. When determining the optimized production process within the production plant 2, the production optimizer 8 takes into account the production-related specifications of the individual plant components 3.

[0093] For example, one or more optimization methods 15 are stored or implemented in the production optimizer 8.

[0094] The determination of the optimized production process within the production plant 2, in particular the at least one optimization method 15, is based, for example, on methods of linear or quadratic programs, genetic optimization, reinforcement learning with Q-tables, neural networks, simulated annealing, Metropolis, swarm algorithms, hill climbing, Lagrange multiplier method or the like.

[0095] The production optimizer 8 accesses at least one prediction model 7 to predict the effects of certain parameter choices. Various mechanisms can be used during optimization. For example, the goal may be to achieve a target variable as accurately as possible or to minimize the probability that a certain variable lies outside the required range. Actuator variables are particularly taken into account here, as they can influence the product and achieve a change in the target variables. During optimization, attention can be paid to constraints so that a concretely implementable set of target specifications is the result of the optimization.

[0096] The production optimizer 8 preferably considers constraints 16. These constraints must be met, for example, to allow the specification of setpoints from a plant or process perspective. This applies, for example, to maximum rates of change of actuator activity and / or the dependencies of an adjustable variable on other adjustable variables.

[0097] The production optimizer 8 expediently uses a target optimization 17. Target optimization 17 is the specifically implemented procedure, which utilizes all of the presented subsystems to finally determine which target specifications are to be used to achieve certain target values. Generally, a product must simultaneously achieve several target values, so either it is ensured that only different parameters are assigned target specifications for the various target criteria, or a joint target optimization must be created for the two or more target criteria, so that a clear target specification is established for the selected parameter. Target optimization is used before production begins to generate the first target specifications.As the process progresses, target specifications for subsequent process steps can also be generated or updated at a later point in time, as long as the product has not yet left the sphere of influence of a specific actuator 11. During this update, a different target optimization can also be used for a given target variable, e.g., because certain sensory data only becomes available during or after a process step but enables a more precise / better target specification for still-available parameters.

[0098] A production plant control system 9 generates target specifications for the plant automation system 4 based on the optimized production process determined by the production optimizer 8.

[0099] The inventive system 1 from Figure 1further comprises a central data storage 14. The other components of the system 1 can access this central data storage 14 if necessary. For example, the production plant 2 or the plant components 3, and in particular the plant automation 4, can store data in this central data storage 14. For example, data from the sensors 10 and / or actuators 11 are stored there. The data stored in the central data storage 14 is used by the model generator 6 to generate the at least one prediction model 7.

[0100] Particularly preferably, the model generator 6 generates a plurality of different prediction models 7. The different prediction models 7 differ, for example, with regard to the creation method, the original data and / or the learning algorithms.

[0101] The process optimizer 8 takes into account the prediction model 7, which provides the currently best prediction for the product currently manufactured in the production plant 2 and / or for the products to be manufactured in the future according to the production planning system 5, in particular taking into account one or more target criteria.

[0102] Furthermore, when determining the optimized production process within the production plant 2, the process optimizer 8 performs an evaluation with regard to one or more target criteria. This is done, for example, by considering the target criteria through the optimization process 15.

[0103] The process optimizer 8 determines, in particular, the effects of setpoint changes on the target specifications of the product currently manufactured in production facility 2 and / or the product to be manufactured in the future according to the production planning system 5. This is, for example, part of target optimization 17.

[0104] Advantageously, the process optimizer 8 includes information on possible setpoint changes of the production plant 2, in particular of the individual plant components 3. The information particularly concerns setpoint changes that can be implemented by the production plant 2, in particular of the individual plant components 3, such as maximum rates of change of actuator activity and / or dependencies of an adjustable variable on other, similarly adjustable variables. This information is part of the constraints 16.

[0105] The invention further relates to a method for controlling a production plant 2 consisting of several plant components 3, in particular a production plant 2 for producing industrial goods such as metallic semi-finished products. The method is implemented, for example, by a system 1 according to Figure 1The system 1 can be implemented at least partially by software executed by a computing device, wherein the computing device is designed to control a production plant 2.

[0106] The method according to the invention comprises the following steps: Collecting information about the products to be manufactured in production plant 2, in particular target criteria for the products to be manufactured, monitoring the production process within production plant 2, in particular within plant sections 3,

[0107] Generating at least one prediction model 7 for the product currently manufactured in the production plant 2 and / or for the products to be manufactured in the future, in particular with regard to compliance with the target criteria for the products currently or in the future, wherein the recorded information of the production plant 2, in particular of the several plant parts 3, is taken into account in the generation of the at least one prediction model 7, Determining an optimized production process within the production plant 2 based on the recorded information about the products to be manufactured in the production plant 2, the monitoring of the production process within the production plant 2, in particular within the plant components 3, the at least one generated prediction model 7 and production-related parameters of the production plant 2, in particular production-related parameters of the individual plant components 3, generating and executing target specifications for the control and / or regulation of the production process within the production plant 2, in particular control and / or regulation of the processes within the plant components 3.

[0108] The monitoring of the production process within the production plant 2, in particular within the plant components 3, is carried out by means of sensors 10. In particular, the monitoring of the production process within the production plant 2, in particular within the plant components 3, is carried out continuously.

[0109] The method according to the invention further comprises the step of storing data in a central data storage device 14, in particular data relating to the acquisition of information about the products to be manufactured in the production plant 2, in particular with target criteria for the products to be manufactured, data relating to the monitoring of the production process within the production plant 2, in particular within the plant parts 3, data relating to the generation of at least one prediction model 7 for the product currently manufactured in the production plant 2 and / or for the products to be manufactured in the future, in particular with regard to compliance with the target criteria for the products to be manufactured currently or in the future, data relating to the determination of an optimized production process within the production plant 2 on the basis of the acquired information about the products to be manufactured in the production plant 2,the monitoring of the production process within the production plant 2, in particular within the plant components 3, of the at least one generated prediction model 7 and production-related parameters of the production plant 2, in particular production-related parameters of the individual plant components 3, and / or data relating to the generation and execution of target specifications for the control and / or regulation of the production process within the production plant 2, in particular the control and / or regulation of the processes within the plant components 3.

[0110] The generation of the at least one prediction model 7 is based, for example, on methods of statistics, machine learning, artificial intelligence or the like.

[0111] The at least one prediction model 7 determines a probability that one or more target criteria lie within a defined range.

[0112] In a particularly advantageous variant, the method according to the invention comprises the step of generating a plurality of mutually different prediction models 7, wherein the different prediction models 7 differ, for example, with regard to the creation method, the original data and / or the learning algorithms.

[0113] When determining the optimized production process within production plant 2, the prediction model 7 is taken into account in the aforementioned variant, which provides the currently best prediction for the product currently manufactured in production plant 2 and / or for the products to be manufactured in the future, in particular taking into account one or more target criteria.

[0114] When determining the optimized production process within Production Plant 2, an assessment is carried out in particular with regard to one or more target criteria.

[0115] The determination of the optimized production process within Production Plant 2 examines the effects of setpoint changes on the target specifications of the product currently manufactured in Production Plant 2 and / or the product to be manufactured in the future.

[0116] Furthermore, the determination of the optimized production process within the production plant 2 can take into account information on possible setpoint changes of the production plant 2, in particular of the individual plant components 3. These are, for example, setpoint changes that can be implemented by the production plant 2, in particular of the individual plant components 3, such as maximum rates of change of actuator activity and / or dependencies of an adjustable variable on other, likewise adjustable variables.

[0117] The production sequence 18 of the production plant 2 from Fig. 1is marked by an arrow, according to which production proceeds sequentially from left to right. Thus, successive production steps are carried out sequentially, with the control according to the invention affecting all sequential production steps.

[0118] Below and independently of the example from Fig. 1An application scenario for the method according to the invention is explained below. It is known that a heat treatment step has a strong influence on the mechanical properties of metal products. This makes it possible to correct certain process deviations in the previous process stages (e.g. melting, casting, hot rolling, pickling, cold rolling). For this purpose, various models 7 are trained which allow a prediction of the mechanical properties or a prediction that these will meet or exceed a certain validity range. Some of the models 7 only use specified variables, such as carbon content, manganese content, hot rolling temperature, coiling temperature, degree of cold forming, etc. Other models 7 can also use reactive measured variables, such as drive power or forces during forming processes or during the movement of a strip.The optimizer 8 uses such a model 7 to define the target specifications for the individual process steps. As soon as a process step is completed (e.g., metallurgical treatment to adjust the chemical analysis), concrete measured values ​​are also available. The same or a new model 7 can then be used to generate updated target specifications within the optimizer 8. If the same model 7 is used as before, or a different one which uses some of the same input parameters as before, in particular those that are target parameters of the process that has just been completed, the optimizer 8 can no longer use the parameters already implemented during the melting process for optimization, but must instead use the values ​​determined by the sensor system 10. This ensures optimal use of the information about the current product piece.The specifications for subsequent process steps are then based on any deviations in the chemical analysis. Similarly, once the hot rolling process is complete, the specifications for subsequent process steps can be adjusted according to the actual hot rolling or coiling temperature, or a different model 7 can be used, which, for example, uses the rolling forces as an input variable. This process is then repeated upon completion of cold rolling, so that a correspondingly improved set of target specifications is available for the process on the annealing line.

[0119] It is important that, firstly, the generation of a large number of models 7 is part of the system 1, that a large number of quality criteria are optimized simultaneously and, thirdly, that a continuous improvement of all sub-processes takes place using the maximum available information. List of reference symbols

[0120] 1System 2Production plant 3Plant component 4Plant automation 5Production planning system 6Model generator 7Prediction model 8Production optimizer 9Production plant control 10Sensor 11Actuator 12Sensor list 13Actuator list 14Data storage 15Optimization method 16Constraints 17Target optimization 18Production sequence

Claims

1. System (1) for controlling a production plant (2) consisting of a plurality of plant parts (3), particularly a production plant (2) for producing industrial goods such as metallic semifinished products, comprising: the production plant (2) consisting of a plurality of plant parts (3), a respective automatic plant unit (4) for monitoring and controlling and / or regulating the process within each of the plant parts (3), wherein the automatic plant units (4) of the plurality of plant parts (3) are independent of one another and no exchange of information takes place between the automatic plant units (4) of the plurality of plant parts (3), a production planning system (5) with data about the products to be produced in the production plant (2), the data comprising target criteria for the products to be produced, a model generator (6) for generating at least one prediction model (7) for the product currently produced in the production plant (2) and / or for products to be produced in the future in accordance with the production planning system (5), wherein the prediction model (7) is arranged to determine a probability that one or more target criteria of the product currently produced in the production plant (2) and / or products to be produced in the future lie within a defined range, wherein the model generator (6) is arranged to take into consideration the results of monitoring the plurality of plant parts (3) when generating the at least one prediction model (7), a production optimiser (8) arranged for determining an optimised production process within the production plant (2) on the basis of the data of the automatic plant units (4), of the production planning system (5) and of the prediction model (7) generated by the model generator (6), wherein the production optimiser (8) is arranged to take into consideration the technical production setpoints of the individual plant parts (3) when determining the optimised production process within the production plant (2), and a production plant control (9) arranged for generating target setpoints for the automatic plant units (4) on the basis of the optimised production process determined by the production optimiser (8).

2. System (1) according to claim 1, wherein the plurality of plant parts (3) comprises respective sensors (10) for detection of product characteristics, process parameters and / or operating states with the plant part (3).

3. System (1) according to claim 1 or claim 2, further comprising a central data memory (14) for the individual components of the system (1), wherein the individual components of the system (1) can access the data and, in particular, can adapt the data.

4. System (1) according to any one of claims 1 to 3, wherein target criteria are selected from thickness, width, length, weight, tensile strength, yield strength, modulus of elasticity, elongation at break, corrosion resistance, presence or number of surface defects of different kinds, number of cracks at the surface or within the material, drop weight tear test results, Charpy results, transition temperatures from ductile fracture to brittle fracture, layer thickness of zinc coating, tube wall thickness, eccentricity, web height, flange height, flange thickness, profile, planarity and the like.

5. System (1) according to any one of claims 1 to 4, wherein the model generator (6) is arranged to generate a plurality of mutually different prediction models (7), wherein the different prediction models (7) differ with respect to, for example, method of creation, original data and / or learning algorithms.

6. System (1) according to claim 5, wherein the production optimiser (8) is arranged to take into consideration the prediction model (7) which delivers the currently best prediction for the product currently produced in the production plant (2) and / or for the products to be produced in the future in accordance with the production planning system (5), particularly with consideration of one or more target criteria.

7. System (1) according to any one of claims 1 to 6, wherein the production optimiser (8) is arranged to carry out an evaluation with respect to one or more target criteria when determining the optimised production process within the production plant (2).

8. System (1) according to any one of claims 1 to 7, wherein the production optimiser (8) is arranged to determine the effects of target value changes on the target specifications of the product currently produced in the production plant (2) and / or the product to be produced in the future in accordance with the production planning system (5).

9. System (1) according to claim 8, wherein the production optimiser (8) comprises data with respect to possible target value changes of the production plant (2), particularly the individual plant parts (3), particularly target value changes which can be realised by the production plant (2), particularly the individual plant parts (3), such as maximum rates of change of actuator activity and / or dependencies of a settable variable on other similarly settable variables.

10. System (1) according to any one of claims 1 to 9, wherein the production optimiser (8) is arranged to optimise the production process within the production plant (2) for a plurality of product sections.

11. Method of controlling a production plant (2) consisting of a plurality of plant parts (3), particularly a production plant (2) for producing industrial goods such as metallic semifinished products, comprising the steps: detecting data with respect to the products to be produced in the production plant (2), the data comprising target criteria for the products to be produced, monitoring the production process within the plant parts (3), wherein automatic plant units (4) of the plurality of plant parts are independent of one another and no exchange of data takes place between the automatic plant units (4) of the plurality of plant parts (3), generating at least one prediction model (7) for the product currently produced in the production plant (2) and / or for products to be produced in the future, wherein the prediction model (7) determines a probability that one or more target criteria of the product currently produced in the production plant (2) and / or products to be produced in the future lie within a defined region, wherein the detected data of the plurality of plant parts (3) are taken into consideration when generating the at least one prediction model (7), determining an optimised production process within the production plant (2) on the basis of the detected data with respect to the products to be produced in the production plant (2), monitoring the production process within the plant parts (3), the at least one generated prediction model (7) and technical production parameters of the individual plant parts (3), generating and executing target setpoints for the control and / or regulation of the processes within the plant parts (3) on the basis of the optimised production process.

12. Method according to claim 11, wherein monitoring of the production process within the production plant (2), particularly within the plant parts (3), is carried out by means of sensors (10), the monitoring or the production process within the production plant (2), particularly within the plant parts (3), preferably being carried out continuously.

13. Method according to claim 11 or claim 12, further comprising the step of storing data in a central data memory (14), particularly data with respect to detection of data about the products to be produced in the production plant (2), particularly with target criteria for the products to be produced, data with respect monitoring of the production process within the production plant (2), particularly within the plant parts (3), data with respect to generation of the at least one prediction model (7) for the product currently produced in the production plant (2) and / or for products to be produced in the future, particularly with respect to maintenance of the target criteria for the products to be produced currently or in the future, data with respect to determination of an optimised production process within the production plant (2) on the basis of the detected data about the products to be produced in the production plant (2), monitoring of the production process within the production plant (2), particularly within the plant parts (3), of the at least one generated prediction model (7) and technical production parameters of the production plant (2), particularly technical production parameters of the individual plant parts (3), and / or data with respect to generation and execution of target presets for the control and / or regulation of the production process within the production plant (2), particularly control and / or regulation of the process within the plant parts (3).

14. Method according to any one of claims 11 to 13, wherein generation of the at least one prediction model (7) is based on statistical methods, methods of machine learning, methods of artificial intelligence or the like.

15. Method according to any one of claims 11 to 14, comprising the step of generating a plurality of mutually different prediction models (7), wherein the different prediction models (7) differ with respect to, for example, methods of creation, original data and / or learning algorithms.

16. Method according to claim 15, wherein determination of the optimised production process within the production plant (2) takes into consideration the prediction model (7) which delivers the currently best prediction or the product currently produced in the production plant (2) and / or for products to be produced in the future, particularly with consideration of one or more target criteria.

17. Method according to any one of claims 11 to 16, wherein an evaluation is carried out with respect to one or more target criteria when determining the optimised production process within the production plant (2).

18. Method according to any one of claims 11 to 17, wherein determination of the optimised production process within the production plant (2) checks the effects of target value changes on the target specifications of the product currently produced in the production plant (2) and / or the product to be produced in the future.

19. Method according to any one of claims 11 to 18, wherein determination of the optimised production process within the production plant (2) takes into consideration data with respect to possible target value changes of the production plant (2), particularly of the individual plant parts (3), particularly target value changes which can be realised by the production plant (2), particularly the individual plant parts (3), such as maximum rates of change of actuator activity and / or dependencies of a settable variable on other equally settable variables.