Method for controlling a process in a machine for processing plastics - Patents.com

Abstract

It provides the potential for even less experienced and trained operators to quickly establish machine settings that will produce "good parts." A method for controlling a process in at least one machine for processing plastics, comprising the steps of: - running a simulation for manufacturing at least one component under generation of a simulation data set (SD) for the geometrical profile and / or material properties of the component, - determining a process image for running the machine at an idealized operating point based on the simulation data set (SD), - generating a Design of Experiments (DoE) matrix, - iteratively simulating the DoE matrix under reduction of the DoE matrix and calculation of the remaining variations of the process parameters to obtain a trained process model for the machine, allowing the component to be manufactured on the machine, - validating the remaining variations of the process parameters by real experiments (40) in which a component is manufactured on the machine and evaluated in order to generate a Process Parameter Data Set (PPD) for the subsequent running of the machine at an operating point (AP), the operator communicating with a software communication robot, in particular a chatbot, the method steps comprising at least two artificial intelligences interacting with each other.

Description

[Technical field]

[0001] (Related Applications) This application is related to and claims priority from German patent application No. 10 2022 102 748.1, filed on February 7, 2022, the disclosure content of which is hereby expressly incorporated in its entirety into this application.

[0002] FIELD OF THEINVENTION The invention relates to a computer-implemented method for controlling a process in at least one machine for processing plastics and other plasticizable materials having the features of claim 1, as well as a corresponding machine control device for carrying out said method having the features of claim 12 and also to a computer program product for carrying out said method having the features of claim 13.

[0003] Before the prior art and the invention which proceeds from it are described in detail below, first some concepts are defined which are used within the framework of the present application.

[0004] If the term "component" is used within the framework of this application, it should not be interpreted restrictively, since the method can also be used in multi-component injection molding, multi-cavity injection molding, injection molding with so-called family tools, or in known special methods or also in additive manufacturing. In all these cases, components, objects and objects are produced, for which the term "component" is used as a synonym.

[0005] If the term "process parameters" is used within the framework of this application, it means the parameters required for producing a component in a plastic processing machine, by means of which such a machine produces the component. Typically, these parameters are, inter alia, pressure, travel distance of the conveying means, temperature, speed, filling volume, discharge volume, drop size, etc.

[0006] Correspondingly, the process parameter data set is the collection of all process parameters that are necessary to describe such a plastic processing machine for manufacturing the component.

[0007] If the term "expert knowledge" is used within the context of this application, it refers to the knowledge available to a skilled operator of such machines. This knowledge can be current or liquid knowledge, but it also includes non-liquid knowledge, which can be obtained, for example, from specialized literature. This knowledge is based in particular on information about the procedures and the relevant situations of the components involved in the processes that run on these machines, and about their interactions with one another. This expertise also includes, for example, knowledge of the respective machines and their machine data or their structure, as well as knowledge of materials and their properties. Expert knowledge enables a trained user to operate such machines, at least as long as the components can be produced and the machines are operable. Expert knowledge also includes, however, all the knowledge that is necessary to produce qualitatively high-value components.

[0008] Within the context of this application, the terms "process model" and "process image" refer to a virtual model or image of a process that is actually possible on a machine, i.e. a method for manufacturing a component. A process model is usually a data set that virtually represents a process that is actually possible. A "base model" is an initial process model.

[0009] A "process window" is a portion of such a process model within which machines processing plastics and other plasticizable materials can be operationally operated using corresponding process parameters or process parameter data sets.

[0010] Within the framework of this application, the term "operating point" means a point that is set by means of a process parameter data set on a machine for manufacturing an object, at which the machine is operationally run.

[0011] When "AI model" is spoken of within the framework of this application, it relates to a model of a process or a model of a procedure, created by means of artificial intelligence (AI). [Background technology]

[0012] (Prior Art) For example, in the automotive sector, the demands on plastic or plastic-like components are increasing. Topics such as lightweight construction, recyclability and design requirements lead to more and more complex tool technology and automation, and therefore to more and more demanding processes. At the same time, due to a general shortage of specialists, it is becoming more and more difficult to obtain suitable operating personnel for complex machines, equipment and processes. The topic of sustainability leads to the use of recycled materials, which, for example as non-homogeneous mixtures, have different and non-reproducible properties than so-called "virgin" materials. These variable properties of plastics result in processes having to be followed and adapted in ongoing operation.

[0013] Rule-based methods reach their limits in this complexity. Also, via sensors inside the machine, it is not always possible to evaluate the component quality based on process changes. Achieving a so-called "good part" ultimately determines the requirements for the component. Parameters such as dimensional accuracy, distortion, sink marks, overmolding, gaps, surface quality, color, weight, etc. are only some aspects that determine the quality and usability of a plastic product.

[0014] In the so-called filling simulation, methods are already known that can simulate the filling profile and the flow front course from the so-called injection point for a given plastic material or material of a given plastic class in an injection molding tool configuration. Such a method is known, for example, from DE 10 2015 107 025 B4.

[0015] It is known, for example from WO 2020 / 058387 A1, that these filling simulations are supplemented with machine-specific algorithms, from which parameters for workshop programming of the injection molding machine, such as speed, pressure and temperature profiles, can be derived. This filling simulation, expanded by the machine action, can be saved as a model for exactly one operating point. This means that already in the first run of the injection molding tool, a releasable component can usually be produced, thereby avoiding tool damage.

[0016] Or the operator can slowly work his way towards a complete filling of the mould cavity by trial and error, i.e. the operator sets the process. In the context of so-called Design of Experiments (DoE) methods, it is optimal to detect the process limits and the dependencies of the machine / process parameters on the component quality. For example, the question can be posed: how does the component change if the holding pressure profile is increased by 5%? How does the component change if the switch-point is moved from injection to holding pressure? On the one hand, rule-based work instructions can be derived from this for the operator, or an automatic adaptation of the process values ​​can be carried out in the next step. If this problem arises, one tries to manipulate the process via a manual or automatic change of parameter A, thereby bringing the product again into the range of good component quality (the process is automatically optimised in the case of variations).

[0017] As already mentioned, rule-based methods reach their limits earlier here due to complexity.

[0018] In the field of so-called "artificial intelligence", various possibilities are known regarding machine learning and conversational communication: Thus, from DE 10 2020 111 128 A1 a machine learning method and monitoring system for additive manufacturing is known. Likewise, a learning unit cooperating with a machine is known, for example, from DE 10 2017 006 703 A1, which discloses a machine learning device for teaching current commands for a motor. An automatic and adaptive production method in robotic production, in which a robot learns the position and orientation of other elements on the basis of geometric relationships, can be found in DE 11 2016 002 431 T5. For interactive contact with such robots, software communication robots, also known as chatbots, are used, which contact operators via text recognition and / or language recognition (see Patent Document 6 (WO 2020 / 020515 A1) below).

[0019] The following Patent Document 7 (US 2021 / 0362242 A1) discloses a rapid material development process for Powder Bed Fusion Additive Manufacturing (PBF AM) processes, which generally uses (computer) computational fluid dynamics (CFD) simulations to predict an orthogonal parameter space with an ideal parameter set, which facilitates the selection of a simulated parameter set that can then be used in a Design of Experiment (DoE) experimental plan to predict an orthogonal parameter space with an ideal parameter set. The orthogonal parameter space defined by DoE can then be used to generate multiple structure samples with reduced volumes (portions of structure samples) using PBF AM with varying laser or electron beam parameters and / or varying feedstock chemistry (composition). The reduced volume structure samples are mechanically characterized and analyzed to provide an optimal parameter set for a 3D article or validation sample, for example, using high-throughput techniques such as X-ray tomography, which allows for a better understanding of the parameters and their effect on defects and microstructure. Additionally, the relationships between input process parameters and material characterization outputs are modeled so that machine learning techniques (artificial intelligence) can be used to optimize future parameter selections, using machine learning techniques in simulations and experimental designs.

[0020] DE 10 2017 130 997 A1 discloses a method for simulating a forming process or a part of a forming process, in which the state of the objects involved in the forming process, in particular the forming machine, the forming tool and / or the material to be processed, is calculated in successive discrete time steps under preset conditions, the conditions representing input parameters of the forming process, in which (a) after a time step located before the end of the simulated forming process or a part of the forming process, an inspection of the calculated state of the objects involved in the forming process is carried out on the basis of at least one quality criterion. (b) If the inspection after step (a) shows that at least one quality criterion is not met, at least one of the following steps is carried out: restart the simulation by repeating the calculation of the time step and / or the preceding time step; continue the simulation by calculating the time step following the time step. (c) In this case, the conditions are at least partially changed during the execution of method step (b). An operator interaction is possible by means of an interface. The method also uses artificial intelligence during the simulation to enable an optimized manufacturing process.

[0021] From US 9 138 929 B2, a method is known for simulating the process of filling a mold and the process of ejecting a part from a mold cavity by means of ejector pins. The method includes simulating the process of filling a mold by providing a three-dimensional computer model defining the geometry of a solution domain, determining boundary conditions, discretizing the solution domain based on the model to form a mesh with a number of cells, combining physical data for the material, solving energy equations and flow equations for at least a portion of the solution domain, and calculating flow and temperature conditions in the cells as a function of time to obtain a mold filling simulation result. The method further includes an ejection simulation simulating the process of ejecting the simulated part from the mold cavity by accepting user input via a user interface to define the number of ejector pins, the positioning of the ejector pins, the geometry of the ejector pins, and the speed at which the ejector pins are moved. The method further includes using results of the mold filling simulation as starting conditions for the ejection simulation, solving static equilibrium equations for at least a portion of the solution domain, and calculating the effect of the ejection process on the simulated part based on the number of injector pins, the positioning of the ejector pins, the geometry of the ejector pins, and the speed at which the ejector pins are moved. The method does not include manufacturing of a real part, does not include evaluation thereof by an operator, and does not include interactive communication by which an operator can mark the part on a three-dimensional representation. [Prior art documents] [Patent documents]

[0022] [Patent Document 1] German Patent Invention No. 102015107025 [Patent Document 2] International Publication No. 2020 / 058387 [Patent Document 3] German Patent Application Publication No. 102020111128 [Patent Document 4] German Patent Application Publication No. 102017006703 [Patent Document 5] German Patent Application Publication No. 112016002431 (WO German translation) [Patent Document 6] International Publication No. 2020 / 020515 [Patent Document 7] U.S. Patent Application Publication No. 2021 / 0362242 [Patent Document 8] German Patent Application Publication No. 102017130997 [Patent Document 9] U.S. Patent No. 9,138,929 Summary of the Invention [Problem to be solved by the invention]

[0023] Starting from this prior art, the object underlying the present invention is to provide even a trained operator with little experience the possibility of quickly setting up a machine which is capable of producing a "good part". [Means for solving the problem]

[0024] The problem is solved by a method having the features of claim 1 as well as by a machine control device having the features of claim 12 and also by a computer program product having the features of claim 13.

[0025] Advantageous further configurations are the subject of the dependent patent claims. The features recited individually in the patent claims can be combined with one another in a technically meaningful manner and can be supplemented by what is described in the present specification and by details from the drawings, which show further implementation variations of the invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0026] Hereinafter, a mode for carrying out the invention will be described.

[0027] A computer-implemented method is used for controlling a process in at least one machine for processing plastics and other plasticizable materials. A simulation for manufacturing at least one component is first performed on the machine, generating simulation data sets, or alternatively, corresponding data sets are loaded, which contain the simulations required for the manufacture of the component, whereby these data sets relate at least to the geometrical contour of the component and / or to the material properties of at least one material to be processed for manufacturing the component.

[0028] These simulation data sets are used to create a process image for operating the machine at idealized operating points. This process image, which has been developed through numerous simulations using artificial intelligence, can be used, so to speak, as a starting model for the virtual operation of the machine.

[0029] The artificial intelligence base model is subsequently further trained and validated by a series of statistical experiments organized in a so-called "design of experiments" (DoE) matrix. The statistical design of experiments using the DoE matrix results in a set of possible combinations of parameter values ​​for operating the machine. By iteratively simulating the DoE matrix, the remaining variations of the process parameters that can realistically (actually) operate the machine are determined. This thereby results in a reduction of the possible combinations of parameter values. The original base model then becomes a trained process model for the machine.

[0030] Now, in the next step, the variations of the process parameters remaining after the simulation of the DoE matrix are verified under the execution of real experiments. In this case, the evaluation of the components manufactured on the machine is carried out by the operator communicating with a software communication robot, in particular a chatbot, which can recognize the text input and / or voice input and / or gesture input by the operator and output a text message and / or voice message and / or gesture to the operator in response. In this case, in the case of operator-machine communication, other visual or acoustic communication methods are also included, and basically also methods via other sensory channels, such as, for example, kinesthetic, olfactory or gustatory. In this case, the above-mentioned steps of the method also include at least two interacting artificial intelligences.

[0031] This results in a process parameter data set for and at an operating point (or operating point) at which the machine can be used to produce a component that is evaluated as good. This processing method has the advantage that, with the use of at least two artificial intelligences, the wide variety of possible process parameter data sets is reduced to a meaningful range. A trained operator can also quickly reach an operating point at which the machine produces a qualitatively high-value component by communicating with a software communication robot, i.e., for example, a chatbot.

[0032] However, a process model is also created at the same time, which advantageously guarantees the further operation of the machine. If, for example, a change in the material charge makes a difference, the machine first automatically changes within certain limits based on the current process model and continues to produce the component. At the same time, if a departure from the process model occurs, the prerequisites are created for the machine or the artificial intelligence to be able to contact the operator if necessary.

[0033] The aim of the development is, on the one hand, to develop an AI basic model of the process by numerous simulations and to further train and validate it by means of a statistical series of experiments (DoE). The new AI process model optimized from this is then coupled with another second AI model, i.e. a software communication robot, in particular a chatbot. Here, in the simplest way, in a dialogue with an operator, preferably with a graphical aid using a 3D image of the plastic component, the current component quality is fed back to the AI ​​process model. At least two AI models interact with each other and the quality of the models can be further developed by machine learning in continuous operation.

[0034] Preferably, in the step in which variations of process parameters are verified under the execution of real experiments, an automated dialogue is carried out with a control device (open-loop control device / closed-loop control device), which advantageously communicates with at least one evaluation system for evaluating components in order to further optimize the process.

[0035] Preferably, the method is carried out on injection molding machines or machines for additively manufacturing components, which are precisely the machines that need to advantageously quickly and smoothly adapt to changing framework conditions (construction conditions) in the field of plastics processing machines, either due to the use of recycled materials or due to the increasing complexity of components.

[0036] As simulations from which simulation data sets are created, very different process simulations come into consideration. Here, for example, it can concern a procedure that takes into account the cooperation of an injection moulding unit and a mould closing unit in relation to the injection and holding pressure phases. It has been found that a particularly preferred embodiment is to use a filling simulation for filling a mould cavity of an injection moulding tool on an injection moulding machine as a simulation for producing at least one component on a machine, since this reflects well the problems in producing the component.

[0037] Preferably, to generate the Design of Experiments (DoE) matrix, the artificial intelligence has access to a database of expert knowledge in which various pieces of information are classified according to classes, for example relating to the following classes: - class of mechanical properties, - Class of material properties, - component class, - Tool class for injection molding tools, - the fill time class in the injection process for manufacturing the component, - The relationship between the flow passages and wall thickness within the components, - Plastic class, - Material class.

[0038] In this case, the classification is preferably carried out on the basis of these criteria in cooperation with artificial intelligence in order to call up the relevant information from the expert knowledge [database] for the calculation and make it accessible to the operator, which has the advantage that information that is important for production can be quickly and reliably extracted from the expert knowledge together with the operator and made available for further use.

[0039] Preferably, the process parameter data sets are calculated taking into account material information that can be predefined by an operator and / or selected from expert knowledge, with which the process model can be rapidly sensitized to sudden material variations so as to be more responsive to manufacturing quality high-value components.

[0040] Advantageously, in order to simplify the evaluation by the operator within the context of the interactive communication with the software communication robot, the component can preferably be represented three-dimensionally on the display unit, so that the operator can mark important points in the component representation for the software communication robot. This can be important, for example, if sink marks or lack of surface quality are found in the component or if dimensional accuracy is not guaranteed. Other evaluation criteria are also known to the person skilled in the art.

[0041] Using the acquired process parameter data set, the machine is operated at an operating point that is a part of the process window. This process window is also a part of the process model that covers and includes all possible operating ranges of the machine. As long as the machine is operated within the process window, the production of the component is possible with the evaluated process model with good results. For this reason, the process window is also monitored whether the process window is left during the operation of the machine. This advantageously makes it possible to ensure reliable operation. In the event of leaving the range, the settings can be changed by interactive contact with the operator so that operation is again possible within the process window, or the process window can be adapted to the changed conditions, especially if leaving the process window still produces good parts.

[0042] Preferably, when leaving the process window, an iteration of the method steps is performed starting from the step of generating a Design of Experiments (DoE) matrix, but at the latest starting from the validation of the remaining variations of the process parameters by real experiments. Advantageously, known method steps allow for targeted checking and / or adapting of the process model and therefore also checking and / or adapting of the process window.

[0043] The method is preferably also suitable for carrying out the method on a number of machines, thereby forming and evaluating clusters, in particular for the same machine configuration, for the same process and / or for the same material to be processed. The thus obtained and increased amount of data allows further sharpening of the artificial intelligence process models. Users of such facilities can spread the thus generated algorithms from their "model factory" to other production locations, everywhere obtaining the same good production quality with simultaneous know-how protection.

[0044] Preferably, the algorithms so generated are made available (in part), at least in excerpt form, to the machine builder, so that such information can advantageously be used to sharpen the base model, i.e. the initial process model.

[0045] The object is also achieved by a machine control device for a machine processing plastics and other plasticizable materials, adapted, implemented and / or configured to carry out the method.

[0046] The object is likewise achieved by a corresponding computer program product, stored on a computer-readable medium, having a program code suitable for carrying out the method.

[0047] Further advantages are evident from the further subclaims and the following description of embodiments.

[0048] The present invention will now be described in detail with reference to the embodiments shown in the accompanying drawings. [Brief description of the drawings]

[0049] [Figure 1] FIG. 2 shows a schematic flow chart of the method. [Diagram 2] FIG. 1 shows a process element diagram. [Diagram 3] FIG. 1 shows a diagram of a process extension. EXAMPLES

[0050] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will now be described in detail using examples in conjunction with the accompanying drawings. However, these examples are merely illustrative and should not be construed as limiting the concept of the invention to a specific device. Before describing the present invention in detail, it is pointed out that the present invention is not limited to each component of the device and each method step of the method, because these components and methods can be modified. Furthermore, the concepts or terms used herein are defined only to describe specific embodiments and are not used as limitations. In addition, when the singular or indefinite article is used in the present specification or claims, the elements may be in the plural unless the overall context clearly indicates otherwise (correspondingly, the singular also represents the plural in Japanese translations).

[0051] Fig. 1 shows a computer-implemented method for controlling a process in at least one machine for processing plastics and other plasticizable materials. In step 100, a simulation is carried out for manufacturing at least one component on the machine, with the generation of a simulation data set or simulation data SD. Alternatively or additionally, these data can be read from simulation data sets or simulation data SD already generated when manufacturing at least one component on the machine according to step 102. These simulation data sets include the simulations required for the manufacture of the component, which simulations are suitable for simulating process steps. Typically, the simulation data sets relate at least to the geometrical contour of the component and / or to the material properties of at least one material to be processed.

[0052] The machines are preferably injection molding machines for processing plastics and other plasticizable materials or machines for additively manufacturing components, for which there is at least information about the operation of the machines in the expert knowledge [database] 120, as well as general machine data MD, but also machine data MD for the specific machine on which the production of the components is to take place. In principle, use in other plastic processing machines is also possible.

[0053] The construction and operation of such injection molding machines for processing plastics and other plasticizable materials, which include an injection molding unit and a mold closing unit with a (injection molding) tool received between tool plates and in which the injection molded part is formed in the mold cavity of the tool, are known to those skilled in the art, and will therefore be described below only to the extent necessary for the understanding of the invention. Likewise, the construction and operation of machines for additively manufacturing components, for example by injecting droplets or lines of material in order to build up the component in a construction space, are also known to those skilled in the art.

[0054] If a simulation data set or simulation data SD is available, a process image for operating the machine at an idealized operating point is determined in step 104. The process image is a virtual image of the process that actually takes place on the machine to manufacture the component.

[0055] Now, using the process image, starting from the idealized operating points, a statistical Design of Experiments (DoE) matrix is ​​created in step 106. Basically, however, it is equally possible that the very first DoE matrix is ​​created solely on the basis of expert knowledge 120, as illustrated in Fig. 2. The DoE matrix contains a set of possible combinations of parameter values ​​for running the machine, e.g. predefined pressure values, temperature values, speed values ​​or waymarks or corresponding courses of time, distance and pressure that may be required for the running of the machine.

[0056] After the creation of the DoE matrix, in step 108, a simulation is performed again in order to make accessible, as far as possible, only those variations of the process parameters of the further processing that allow the machine to be operated realistically. This is done by iteratively simulating the DoE matrix under calculation and determination of the remaining variations of the process parameters, thereby reducing the (number of) possible combinations of parameter values ​​of the DoE matrix. This simultaneously results in the generation of a basic model BM trained by the simulation and thus a process model PM for the machine that allows the manufacture of the component on the machine.

[0057] Now, in the next step 110, the remaining variations of the process parameters are verified under the execution of real experiments (trial or test runs) 40. In these real experiments 40, components are manufactured on the machine and evaluated in step 112, which can result in the generation of a process parameter data set PPD for the machine running at the operating point AP.

[0058] This is done with the intervention of a second artificial intelligence. The operator 20 can communicate interactively with the artificial intelligence in step 112a via a software communication robot CB, in particular a chatbot. The software communication robot CB recognizes the voice input and / or text input and / or gesture input by the operator 20 and outputs information on the operating state or the state of the components of the machine as voice and / or text and / or gesture or displays said information. To this end, the software communication robot CB is connected to a control device for data communication (open-loop control device / closed-loop control device) for open-loop and / or closed-loop control of the machine based on the voice input and / or text input and / or gesture input recognized by the software communication robot CB.

[0059] Preferably, alternatively or also supplementarily, a downstream process automatically evaluates the component in step 112b by means of an automated dialogue with a control device (open-loop control device / closed-loop control device) communicating with at least one evaluation system for evaluating the component, for example a corresponding automatic weighing (Waagen) or optical measuring device is conceivable.

[0060] The result of the evaluation, and thus the result of the method steps so far, is a process parameter data set PPD at an operating point AP at which the machine produces a good part as a component. The machine is then operated at this operating point AP.

[0061] In other words, the process on the machine / equipment is guided using multiple AI models generated by ML (machine learning), where the number of models is not limited to two. However, in the following, the method is described based on two AI models for a more understandable explanation. However, further AI, for example, performing further simulations to obtain further data or communicating with other machines or machine parts, may be considered to further support the desired result of reliable manufacturing of the component.

[0062] A first AI model is derived from the filling simulation, e.g. as a base model BM. From this, e.g. idealized operating points are also derived. On the machine, depending on the machine technology, process parameters are specified with the aid of AI for the DoE procedure. This model depends on the components, the process type, the machine properties and / or the material properties of the material to be processed, e.g. plastic or plastic class (plastic-like), and provides for meaningful variations of the process parameters. Variations within a plastic class are also taken into account here.

[0063] This DoE data set, usually a DoE matrix, is reflected to the simulation either directly on the machine or via a digital interface, on the edge or in the cloud. If the performance of the machine controller 10 on the machine is sufficient, then a number of simulations can be calculated directly from this DoE data set, simulating parameter variations in different dependencies. The result is an extended process model trained for each simulation, which is now provided to the machine as a new model. If the simulation cannot be performed on the machine controller of the machine, this simulation is performed externally on the edge or in the cloud. The new extended model of the first AI is then validated with real experiments 40 in a real DoE and additionally sharpened (machine learning). This means that in this dialogue, the results "component good" or "component bad, because ..." must then be fed back to the model of the first AI in order to sharpen it.

[0064] Now, for this purpose a second AI is used, which makes it possible to provide the operator 20 with the necessary information (data) in a simple dialogue via the operator interface UI. This second AI is implemented as a software communication robot, i.e. for example as a chatbot, which "pulls out" the necessary data in the simplest form from the operator 20, preferably using graphical 3D assistance functions.

[0065] An example for an example dialogue in a DoE procedure could be the following dialogue: Machine: "I changed the process parameters. Is the component quality as good as before?" Operator: "No." Machine: "Which quality characteristic is not being met?" Operator: "The component has sink marks and is not completely filled." Machine: "Please indicate the location of the sink mark in the 3D component image." With the 3D visualization the machine shows a 3D image of the (plastic) component to be produced. The operator 20 marks the positions of the sink marks in the graphics directly on the 3D image and confirms this in a dialog. Machine: "Thank you very much. Now please show me which components are not fully filled." Again, the operator 20 indicates to the chatbot CB, i.e. the second AI, the location of the error in the 3D image.

[0066] This is a simple example illustrating a dialogue between the operator 20 and the second AI. This dialogue can be conducted in a chat window via an input device (such as a touch panel) as an operator interface UI, but can also be conducted using voice input.

[0067] In interaction with the preferably implemented 3D visualization and / or with assistance via graphic selection symbols, simple communication with the machine is possible even for a trained operator.

[0068] The first AI model is trained (learned) or additionally sensitized (machine learning) via process information acquired by the second AI. Once this process is complete, the optimized process AI can be activated.

[0069] It has been found to be advantageous to carry out the simulation for manufacturing at least one component on a machine in step 100 as a filling simulation for filling a mould cavity of an injection moulding tool on an injection moulding machine. The filling simulation most quickly reveals where problems may occur in the manufacturing process, for example when larger cross sections alternate with thin-walled cross sections in the moulded part. Furthermore, the filling simulation also reveals where the material is at any point in time. In the filling simulation, it is possible to well recognise whether original virgin material or recycled material is (should be) used, based on the material properties.

[0070] To generate the statistical DoE matrix, preferably expert knowledge [database] 120 is used, which may be available to trained users as Expert Exp. Expert knowledge includes knowledge available in the literature. Expert knowledge also includes knowledge about machines and machine data MD, or about materials and their properties. Expert knowledge can be pre-set for the machine, or it can be trained (learned) for the machine through various process steps, also through the steps according to the invention.

[0071] In this case, the information from the expert knowledge is preferably divided into classes, which are distinguishable, and the classes contain in particular the following information: - class of mechanical properties, - Class of material properties, - component class, - Tool class for injection molding tools, - the fill time class in the injection process for manufacturing the component, - The relationship between the flow passages and wall thickness within the components, - Plastic class, - Material class.

[0072] The classification is carried out by a software communication robot CB (chatbot) based on those criteria, preferably interactively with the operator 20, in order to make the relevant information available for calculation and accessible from the expert knowledge 120.

[0073] As further information for the machine and artificial intelligence, machine data and / or material information of the material to be processed can also be preset by the operator 20 and / or selected as information from the expert knowledge 120. These material information are then used for the simulation, whereby the process parameter data set PPD is calculated taking these information into account. The material information can be current information obtained for the respective charge, which is made available (provided) to the method by a material survey in the individual case on site.

[0074] To accelerate the evaluation method, especially in the interactive evaluation in step 112a, a three-dimensional representation of the component as a 3D graphic 122 has proven to be useful. This means that the operator 20 can observe the component on a display device configured as a user interface UI and mark the positions of the component that the operator 20 would like to evaluate accordingly. This information can then be evaluated by the software communication robot CB and processed accordingly. This approach makes it easier to provide the operator 20 with clear information about where in the component a problem is possibly identified.

[0075] The process model PM obtained by the method can not only be used for the initial setting data set or for the process parameter data set PPD. If anomalies occur during the continued operation of the machine, the machine monitors whether production is still possible meaningfully based on the information of the process model PM present for the machine. By means of the method steps with the simulation of the DoE matrix and the evaluated real DoE matrix, the process model PM is thus present, which is illustrated by a closed frame in FIG. 1 and shows very clearly where and in which framework conditions the operation of the machine is possible. The operation of the machine with the process parameter data set PPD at the detected operating point AP is a part in the process window PF in the process model PM, which is illustrated by a dashed line in FIG. 1. As long as the production of the component takes place with the process parameter data set PPD in the process window PF, no recalculation or intervention of the machine or the machine control 10 is necessary. The machine can be operated further according to step 116. However, if the process window PF is left, readjustment may be necessary.

[0076] If the process window PF is to be left, the machine can first output a warning message. This can be done within the framework of the question 114. However, the machine can also check, either at the command of the operator 20 or automatically on the basis of the information now present in the machine, whether the originally detected process window PF is still correct (valid). It is possible, for example, that in the meantime the material quality has changed due to other charges or that the process parameters are no longer as usable throughout the corresponding run. In this case, the detection of the process window can be carried out anew, preferably starting from step 106 of the generation of a design of experiment (DoE) matrix, but at the latest from step 110 of the validation of the remaining variations of the process parameters.

[0077] Now, in ongoing operation, the process can always be kept within the process window PF that provides the best possible component quality.

[0078] If further irregularities still occur, the process model PM can always be extended and additionally sensitized (machine learning) via the example dialogue described above.

[0079] Parts of the dialogue can of course also be automated, i.e. quality parameters, e.g. the geometry of a component, can be evaluated via a downstream optical inspection system and provided to the model via a standardized digital dialogue. The same applies to other quality parameters such as weight, surface, color, strength, temperature, etc.

[0080] It is also possible that parts of this information can be obtained from downstream processes, for example during inspection in downstream quality assurance, to provide the machine with a learning process in a digital dialogue.

[0081] Ideally, in a relatively large injection molding operation, data from similar or the same machines operating similar or the same processes with materials of similar or the same material class can preferably be "collected". This aggregation "on edge" and evaluation of all machine model data allows to generate better algorithms in the context of federated learning and thus better AI models. Operators can thus gain access to swarm intelligence. The facility operators can spread (deploy) the algorithms so generated from their "model factory" to other production sites and obtain the same good product quality worldwide with simultaneous know-how protection.

[0082] Fig. 2 shows an overview of how the process elements interact. As described in step 100, firstly simulation data SD are provided or used to simulate the production of a component (Fig. 2, top right). This can be done on the machine control 10 or, as shown in Fig. 2, on the edge or in the cloud. This usually results in an initial process image with an initial DoE evaluated via simulation, which creates training data for the base model BM.

[0083] This base model BM (step 104) is then available for use by the machine control device 10 and / or for a simulation computer and thus an AI model.

[0084] If the basic model BM does not yet exist, an experimental procedure DoE is designed with the help of the expert knowledge 120 made available from the expert Exp for the machine, or entered or installed in the machine control device 10, or trained by the process procedure 120, so that the basic model BM is made available to the AI ​​model. However, the expert knowledge can also be involved already in the very first creation of the DoE matrix.

[0085] A machine-specific process model PM is generated from the base model BM by process experiments, i.e. firstly by virtual experiments. This is done in a first step 108 by simulation of the DoE matrix under reduction of possible variations of the process parameters. Starting from the base model, a reduced DoE is provided by simulation, simulated and evaluated, and the base model is sensitized using the training data thus obtained. The DoE matrix is ​​thereby reduced by the base model. This reduced DoE matrix is ​​used as a basis (starting data) for further process experiments. In parallel, the reduced DoE matrix is ​​simulated on the edge / cloud and the base model BM is additionally trained and provided to the control device.

[0086] As a complement, using the further reduced DoE matrix, a real experiment 40 is performed in step 110, whereby process actual values ​​are determined and these process actual values ​​are made available to the process model PM, i.e., based on the DoE matrix, the component evaluations and the process actual values, starting from the base model BM, the process model PM is trained (the base model is used as the basis for the process model).

[0087] In order to also qualitatively evaluate these results, i.e. the desired process parameter data set PPD for the operating point AP, an evaluation EB of the component thus manufactured is carried out. This can be done interactively with the operator 20 (step 112a), by the further AI entering into communication with the operator 20, this being done via the chatbot CB. The chatbot CB can derive analysis and operational recommendations on the basis thereof and provide these to the operator 20. This can also be done with the support of a 3D graphic 122 of the component, as described above. Alternatively, the quality monitoring can also be performed automatically (step 112b) and the evaluation results can be provided. In this case, the chatbot is supported in its function by the expert knowledge 120. In this case, the evaluation of the component can also be carried out in two stages: once during the experiment and subsequently during the productive process, i.e. during the production of the component in the manufacturing process.

[0088] The evaluation thus achieved also allows the second AI to perform a new evaluation of the DoE matrices in order to improve the process model PM, which in turn has an impact on the process parameter data set PPD for the operating point AP as well as on the process window PF within which the machine can reliably produce good components. The improved process model PM can subsequently be used to calculate a new process parameter data set PPD for the operating point AP.

[0089] Furthermore, the machine manufacturer is interested in the machine users providing at least excerpts of algorithms (extracts) generated "on-edge" in the machine manufacturer's cloud. The machine manufacturer collects algorithm excerpts from as many machine users (customers) as possible, i.e. compares these with the content provided by the customers, and thereby sensitizes the generation of the basic model (deep learning).

[0090] This is explained in more detail in the diagram of FIG. 3. The upper part of FIG. 3 shows a cloud solution / edge solution. In this area, as explained with respect to FIG. 2, a simulation and evaluation of the DoE matrix is ​​performed in order to obtain training data for a base model, here a customer-specific base model. The customer can provide this customer-specific base model to the control devices of various machines, such as the machine control device 10 in FIG. 2, each of which is assigned to at least one machine. Via these control devices, as in FIG. 2, data is returned to the simulation, whereby the DoE matrix is ​​improved / reduced and the base model is sensitized. The diagram (scheme) thus far describes a customer-specific on-edge solution.

[0091] But also the results of the evaluated DoE matrices, i.e. for example the training data, are made available at least in the form of extracts (partially) for a cross-customer (widespread) base model, which is created for example at the machine manufacturer, which collects these training data from as many machine users as possible in order to further improve the generation of the base model.

[0092] The advantages cited with respect to the present method may also be obtained in a machine control device 10 for machines that process plastics and other plasticizable materials, so long as the machine control device 10 is adapted, implemented, and / or configured to properly perform the present method.

[0093] Similarly, the method may be advantageously achieved through the use of a computer program product having program code stored on a computer readable medium, whereby the method may be performed using the program code.

[0094] Obviously, the present description is susceptible to widely different modifications, variations and adaptations, provided they fall within the scope and range of equivalence of the appended claims. [Explanation of symbols]

[0095] 10 Machine control device 20 Operator 40 Experiments 100 steps 102 Steps 104 Steps 106 Steps 108 Steps 110 Steps 112 Steps 112a Step 112b Step 114 questions View Template 116 Steps 120 Expertise [Database] 122 3D graphics AP operating point (operating point Arbeitspunkt) BM Basic Model CB Software Communication Robot (Chatbot) DoE (Design of Experiment) EB evaluation components MD Machine Data PF Process Window PM Process Model PPD Process Parameter Data Set SD Simulation data or Simulation data set

Claims

1. 1. A computer-implemented method for controlling a process in at least one machine for processing plastics and other plasticizable materials, comprising: The method comprises the following steps: a) performing a simulation for the manufacture of at least one component on a machine, generating a simulation data set (SD) and / or reading a simulation data set (SD) generated during the manufacture of at least one component on a machine and containing a simulation required for the manufacture of the component, the simulation data set (SD) relating at least to the geometrical contour of the component and / or to the material properties of at least one material to be processed, b. determining a process image for operating the machine at an idealized operating point based on the simulation data set (SD); c. Starting from the idealized operating point, generating a Design of Experiments matrix (hereinafter referred to as a "DoE matrix"), where the DoE matrix includes a set of possible combinations of parameter values ​​for operating the machine; d. Iteratively simulating the DoE matrix under calculation and determination of remaining variations of the process parameters that will realistically allow the machine to be run, where the possible combinations of parameter values ​​of the DoE matrix are reduced, and a process model for the machine is generated that is trained by the simulation and that will allow the machine to manufacture the component; e. validating the remaining variations of the process parameters under execution of real experiments (40) in which components are manufactured and evaluated on the machine to generate a process parameter data set (PPD) for running the machine at an operating point (AP) and subsequently running the machine with the process parameter data set (PPD) at the operating point (AP); The configuration includes: An operator communicates with a software communication robot (CB), in particular a chatbot, and the software communication robot (CB) recognizes a voice input, and / or a text input, and / or a gesture input by the operator, and / or outputs or displays information about the operating state of the machine or information about the state of components, and the software communication robot (CB) is connected to an open-loop control device / closed-loop control device for data communication in order to open-loop control and / or closed-loop control of the machine based on the voice input, and / or the text input, and / or the gesture input recognized by the software communication robot (CB); and Steps a to e include at least two artificial intelligences (PM, BM, CB) interacting with each other; The method comprising:

2. In step e, an automated dialogue is carried out with an open-loop / closed-loop controller communicating with at least one evaluation system for evaluating components; The method according to claim 1 , characterized in that

3. the method is carried out in an injection molding machine or in a machine for additively manufacturing components, 3. The method according to claim 1 or 2, characterized in that

4. the simulation for manufacturing at least one component on a machine is a filling simulation for filling a mold cavity of an injection molding tool on an injection molding machine; The method according to any one of claims 1 to 3, characterized in that

5. To generate the DoE matrix, information from the expert knowledge (120) is used, which information is classified and distinguishable by classes according to at least one of the following criteria: - mechanical property class, - material property class, – component class, - Tool class for injection molding tools, - the filling time class in the injection process for manufacturing the component, - Relationship between flow passages and wall thickness within the component, - Plastic class, - material class, However, the software communication robot (CB) performs the classification based on these criteria, preferably interactively with the operator, in order to make the relevant information accessible from the expert knowledge [database] (120) for the calculation; The method according to any one of claims 1 to 4, characterized in that

6. the machine further includes, as information, material information of the material to be processed, which is preset by the operator and / or selected as information from the expert knowledge (120), which material information is used for the simulation, and the process parameter data set (PPD) is calculated taking into account this material information, The method according to any one of claims 1 to 5, characterized in that

7. In interactive communication between the software communication robot (CB) and an operator, the component is displayed in three dimensions, and at least a portion of the component can be marked by the operator within the display for communication with the software communication robot (CB); The method according to any one of claims 1 to 6, characterized in that

8. Operation of the machine using a process parameter data set (PPD) at an operating point (AP) is part of a process window (PF) in a process model (PM), operation of the machine within the process window (PF) enables production of a component, and operation of the machine is monitored by a machine controller (10) for leaving the process window (PF); The method according to any one of claims 1 to 7, characterized in that

9. - when leaving the process window (PF), the method according to claim 1 is preferably executed anew, starting from the step of generation of a DoE matrix and at the latest from the validation of the remaining variations of the process parameters, The method according to claim 8, characterized in that

10. The method is performed on a plurality of machines, from a respective process model (PM) trained by simulation, and / or from the respective generated process parameter data sets (PPD) and / or - From each operating point (AP), Clusters are constructed with the same or similar machine configurations, with the same process, and / or with the same or similar materials to be processed, and these clusters are evaluated by federated learning to run machines in a process model (PM) that is adapted according to the evaluation; The method according to any one of claims 1 to 9, characterized in that

11. The method includes the steps of: - comparing, at least in part, the results of the evaluation with one another, preferably on a cloud-based basis, and detecting a comparison result; - sensitizing the process models (PM) required for the method based on the results of these comparisons; containing, The method according to claim 10, characterized in that

12. A machine control device (10) for a machine for processing plastics and other plasticizable materials, comprising: said machine control device (10) being adapted, implemented and / or configured to carry out the method according to any one of claims 1 to 11; A machine control device comprising:

13. 1. A computer program product having program code, comprising: said program code being stored on a computer readable medium for carrying out the method according to any one of claims 1 to 11; 1. A computer program product comprising:

Images