WO2002014967A2 - Method for carrying out an automated production process - Google Patents

Abstract

The invention relates to a method for carrying out an automated production process, wherein the production process is controlled based on the measurement of physical or chemical parameters of the product and/or the production processes and of disturbance values (product/process parameters). An analysis is made on which parameters characterize the quality and/or the quantity of the product and/or of the production process in a sufficiently exact manner and which adjustable or non-adjustable but variable product/process parameters influence the characterizing variables (variables of influence). The evaluation of deliberate or random changes in the variables of influence serves to detect the mutual dependency of the variables of influence and this dependency is saved as a dependency model (neuronal network). The dependency model allows calculation and adjustment of the optimum characterizing values and the pertaining adjustable variables of influence. The dependency model (neuronal network) is continuously developed in an automatic manner by evaluation of the process data derived from production.

Description

 Process for carrying out an automated production process

The invention relates to a method for carrying out an automated production process according to the preamble of claim 1.

In every industrialized production process there are two overlapping components: the process itself as the entirety of the mechanical, electronic and informational instruments and the person as the management system that controls the production.

The task of controlling a production facility can be more or less difficult, depending on the complexity of the respective process and the number of variables involved. The task can take on highly critical aspects when it comes to particularly sophisticated automated systems. This fact is of particular importance. Because the automation of production processes is one of the strategic goals that is of the utmost importance for modern industry.

Practically all companies have started this attempt to automate in order to adapt quickly and competitively to the market, a market that requires highly diversified (different) products at low prices and with ever shorter innovation cycles (new editions).

The results of this attempt at automation are not positive in all areas, but they have confirmed, particularly in the so-called "mature" sectors (in which an end product of technological development seems to have been reached for the time being) that automation is an indispensable factor for Represents innovation, especially as far as innovation concerns the production process.

The automation of production processes has shifted the functions previously performed by humans to robots. This is fundamental Changes in the traditional work functions and in the technological structure of the production plant occurred.

However, automation has its limits due to the complexity that arises. When automation is at a certain level of complexity, control and flexibility are lost and the benefits diminish. This also applies to automated production processes in which a large number of physical or chemical parameters of the product or the production process as well as disturbances - more or less - are measured continuously and used in a control loop to control the production process.

While the development of automation has so far been aimed at creating increasingly complex and sophisticated process units, it has been neglected that the previous human function of the worker (operator) can be replaced, but not the function of planning, development, organization, control and control (Management) of the production process. Production requirements have increased and become more and more complex, while compared to the increased production capacity, operators still follow the same principles and decision criteria, terms and models.

The modern systems of the generation and visualization of process data with the help of the new technologies in the automatic control were of no use in the production management, at best these data were ignored.

Therefore, there is a need to synchronize the operational management with the automation and to renew the methods for the management by the operational staff by providing the staff with sophisticated tools to master the complexity of modern, automated production processes. ^It is the object of the invention to design the method for carrying out a lutomatized production process in such a way that the method does justice to the changed role of humans.

ABy), the invention is intended to eliminate the lack of innovative development, ler between the high level of automation technology, on the one hand and still conventional system the aid of learning is, which are applied by the äetriebsleuten in making decisions for control and command of the ^s roduktionsprozesses.

) The solution to this problem results from claim 1.

This creates a tool that connects the one hand between the increasingly sophisticated production process and the

: complex processes and on the other hand creates the person who with

³ planning, development, organization, control and management of the

^P roduktionsprozesses is concerned. The large number of production parameters and their mutual dependency and impact is presented in a manageable and so transparent manner that it is always possible to predict which

Effect an intervention in the process will have.

In this application, ms production parameters and parameters

Influencing variables referred to.

Kenngrößen, die das Ergebnis desFirst of all, a distinction must be made between the influencing factors that need to be controlled and controlled, and

Parameters that are the result of the

/ experience, i.e. in the broadest sense: the quality of the process and / or the

Rewrite ³ products,

The parameters are physical or chemical

'parameters of the product and / or the production process, which determine the desired quality and / or quantity of the product and / or

'' characterize the production process with sufficient accuracy. (Product or

³ process parameters),

On the one hand, the production parameters are physical

> the chemical parameter of the product and on the other hand physical and chemical parameters of the production process and / or disturbances of the

³ products or the production process. It can be seen that only variable influencing variables and only influencing variables that have a previously determined significant and preferably linearly representable influence on at least one of the selected parameters come into consideration.

The influencing variables also include the parameters of the raw materials and auxiliary materials. These parameters as well as the process parameters have to be considered. It must be determined whether the properties of the raw or auxiliary materials vary so widely that ongoing monitoring and inclusion in the virtual process is required. However, the properties and parameters of raw materials and auxiliary materials can also be included in the virtual process if there is enough scope for product planning and - or production planning - that the real process can only be adapted to the product that is actually desired changing raw materials or auxiliary materials is possible.

It should be emphasized at this point that special attention must also be paid to the determination of suitable parameters in the context of this invention. Decisive for such parameters - regardless of whether they are process properties or product properties - must always be the appreciation of the product by the customer of the product (consumer, user, processor). A selection must already be made here in order to achieve the most accurate possible prediction for the desired results on the basis of the selected influencing variables, and in order to enable an accurate statement about the quality of the process and / or the product based on the selected parameters.

From the multitude of process parameters or product parameters, those at most three parameters will be selected which are as reliable indicators as possible that the process works in the target area or the product is produced with the target properties. In the development of the invention according to claim 2, the comparison of the influencing variables or parameters of the virtual model and the measured influencing variables and / or parameters of the real production process (measurement data) is used not only for forecasting but also in the Intervene process to fix systematic errors. As a result, the production of rejects can be avoided.

With the signaling of the prediction result, however, quality monitoring is preferably carried out in the sense that the resulting products are given a classification which characterizes the quality level or the committee without the otherwise usual testing of the individual product or a sample of the product. A quality check and classification of the finished product, which hampers the material flow and is labor intensive, is not necessary.

In the development of the method according to claim 3, comparison of the influencing variables or parameter of the virtual model and the measured influencing variables and / or parameters of the real production process (measurement data) is additionally used to determine faults in the process and / or the informative value of the model for the Process constantly verified. While the size of the deviations detected when the warning signal is output make it necessary to check the model and the process and the interdependency, a pronounced course of the deviations over time (trend) is an indication that a malfunction has occurred within the real process Remedial action is required if the previously defined trust range, which is specific to the relevant influencing variable or characteristic variable, is left.

In the development of the method according to claim 4 and δ, comparison of the influencing variables or parameters of the virtual mode and the measured influencing variables and / or parameters of the real production process (measurement data) is used in order to intervene in a controlling or regulating manner in the process. The comparison of the process data with the data of the virtual process has the advantage that individual measurement data are not used for the intervention in the process, but that the respective controlling intervention takes into account all effects of the intervention due to the stored dependency model. In this way, negative effects of the controlling intervention can be avoided. Inadmissible regulation fluctuations and vibrations are avoided with regulation. The proximity of the process model to the real process is achieved according to claims 6 and 7.

By claim 6, a novel process control is created, which is composed of several steps and with which from the multitude of process parameters or product parameters those at most three parameters are selected which are as reliable indicators as possible that the process works in the soil area, or the product is manufactured with the required properties. It is particularly important to first determine the type and number of trials and test series required so that statistically reliable statements about the dependencies can be obtained. The selection of the production parameters (parameters and influencing variables) that go into the model must also be determined according to whether suitable sensors are available.

However, it is also necessary for the selection that it was determined beforehand that the selected parameter with sufficient probability or reliability makes a statement that the entire process with all its parameters and influencing variables runs stably within a desired target range. Whether a contemplated process parameter meets these requirements has to be determined beforehand through series of measurements and evaluation of the measured values using mathematical and statistical methods. If the parameters are product properties, then depending on the type of property, an ongoing measurement is also possible here; otherwise a measurement in time intervals can also be considered. The further problem here is to prepare the measured variables of the selected influencing and characteristic variables, which arise in tests and or during operation and which are initially only in the dimension of simple physical variables, in such a way that they gain the greatest possible informative value ).

For this purpose, statistical and / or mathematical methods are available today, particularly in the scientific discipline known as chemometry.

It is then determined which dependencies exist between the influencing variables and the parameters. Claim 9 provides suitable aids for determining and representing the mutual dependencies between the influencing variables and the

Effect of the influencing variables on the parameters and on the development of the

Dependency models provided.

These dependencies can be physical / mathematical.

The dependencies can also be such that they cannot be grasped mathematically; in this case the dependencies are determined by experiment. in particular, it will be found that between the influencing variables and the

A two -, three - and more - dimensional dependency exists in the form of a complex matrix of influences and results, which cannot be represented and which does not allow a prediction for product and process »and in particular a control of the process.

It is therefore the intensity of the impact of the influencing variables on the

Characteristics determined. Preferably only the parameters in the

Include consideration that has a significant impact.

Claim 5 provides a suitable method for this.

In any case, this results in a manageable network of dependencies for the operating point or the operating range of the process.

The resulting dependency model can be formulated mathematically, preferably linearly (claim 10) or as a neural network (claim

11) be formulated.

Depending on the complexity of the process, the dependency model is unique and fixed, i.e. it is not changed.

However, since the invention is used especially for complex production processes, as are normal for today's technology, and new perspectives of

Decision making is made possible by claim 2 in further training of

Invention proposed that the dependency model is continuously corrected, supplemented or otherwise updated and stored in any other way by evaluating the process data which is produced continuously or temporarily in production. This update takes place without the

Operator notices this or is bothered by it.

A reliable but manageable model that is suitable for electronic data processing is preferably produced by the method according to one of claims 12 to 14. In this case, mutually dependent production parameters (parameters which are a measure of the quality and / or quantity of the product or the production process or parameters which influence the parameters which can be achieved in the production process) are combined Substitute parameters returned. (Principal Components). In this way, the network of dependencies can be traced back from a multi-dimensional coordinate system to a two-dimensional or three-dimensional dependency without loss of meaningfulness and sensitivity of the model.

In the development according to claim 15, the selected influencing variables and / or characteristic variables or their substitute variables are compared with the corresponding influencing variables and / or characteristic variables or the substitute variables transformed therefrom measured in the real process and the mutual dependencies of the influencing variables are updated. This update can take place continuously, but preferably takes place at selected intervals in order to rule out that systematic errors in the production process or the measurement process are included in the update.

The advantage of the invention lies in the fact that the model of the process to be carried out, including the product, is continuously updated with the actually running process and is described by only a few influences and parameters with previously defined and preferably continuously updated interdependency. It is therefore possible to include the process directly in product planning and production planning and to adapt it accordingly; in other words: in the computer that controls the real process, a virtual process runs simultaneously, taking into account the previously selected influencing variables and parameters. There is a constant comparison between the real process situation and the virtual process situation. This means that particular attention must be paid to the selection of suitable sensors that are able to reproduce the real process data, ie influencing variables and parameters. and on the other hand the real process is controlled as a function of the setpoints predetermined by the virtual process. In the following the invention is illustrated by examples and by means of the figures

1 to 8 further described.

Production processes in this invention are:

❖ Continuous processes that are typical for the petrochemical, chemical or similar industries.

❖ Discontinuous or discrete processes that are used in particular in mechanical engineering, vehicle construction and similar industries.

❖ Mixed processes, typically e.g. B. for the chemical-textiien sector, where continuous and discontinuous processes follow in two different process stages of the same industrial production and the weighting of each stage can be very different.

A production process, through which raw and auxiliary materials are converted into a desired end product, can be represented by a schematic model which, based on the set physical or chemical parameters of the product and / or the production process, the disturbance variables (such as air humidity, Air temperature, air pressure, electrical voltage and the like) and is represented by the monitored parameters (e.g. temperatures, pressures). The control of this process is usually done by manually or automatically filing a certain number of variable parameters that are believed to represent the production process and are critical to the production process and / or the product. It is precisely these variable parameters that are appropriately visualized and statistically evaluated, e.g. through alarm signals, averaging, trend formation, etc., should allow the operator to master the control of the process and consequently also the possible variations of the end product and thereby adapt to the changing market requirements.

Such a production process with significant influencing factors is shown in Fig.1.

The dependencies between the parameters of process and product and the

Setting variables and non-adjustable parameters (i.e. disturbance variables), logistics

(Raw and auxiliary material including electrical, pneumatic energy and water) are visible and give the possibilities of measurement, control and Control of production and product by measuring meaningful parameters (parameters); however, these dependencies can only be represented incompletely in complex processes, so that the possibilities for control are also limited.

Production systems of this type can today be represented as virtual systems in an orthogonal coordinate system with the aid of "Computer Integrated Manufacturing" (C1M).

At the lowest level, as shown in Fig. 2, all elements of the process, that is, for. B. machines, reactors, sensors, controls. The higher levels 1 to 4 contain other functional areas with the peculiarity that they contain increasingly complex reactions and correspond to different or uniformly increasing reaction times. Each area of an automatic control to be executed is to be regarded as an autonomous functional area at the respective level of the CIM, intended to be fully integrated as part of a uniform automation system. According to the basic principle of the CIM, planning is carried out from top to bottom and implementation from bottom to top, so that the automation develops in steps, but in a uniform framework, i.e. H. can be planned and carried out.

In Fig. 3, a plant for the production of continuous, synthetic fibers is shown in a diagram with the process stages: storage of the polymer chips, drying of the polymer, spinning, winding, automatic transport of the bobbins, intermediate storage, selection and packaging (not shown). The process therefore consists on the one hand of a continuous and on the other hand of a discontinuous process, the weighting of both process components being 50:50. In the functional representation of this process according to the GIM principles, only the first two levels are present (represented). The levels of the continuous process are shown on the left and the levels of the discontinuous process on the right. The discontinuous process relates in particular to the automated handling of the product (the Do the washing up). A further subdivision with the subsections was made on level 1 in terms of engineering considerations of the functions

❖ automation,

❖ management and control,

❖ Monitoring.

If one looks at this diagram from the point of view of the operator, the representation appears to be relatively meaningless. Only data, but no information, are available, the data being the values signaled point by point by the display devices, e.g. As temperature, the sum of interrelated data can be referred to as information.

The data that accrue in level 1 of industrial practice can also represent some typical information, such as temporal data, trends, average values, extreme values; however, this information is not sufficient to identify and pursue the correct process strategy. So far, the data and the information compiled from it have only been used to generate an alarm signal and to trigger the necessary measures to deal with the alarm (management and alarm). It is clear here that the functions shown in FIG. 2 on the higher levels II, III and IV, namely in particular:

Product planning at the level of

Corporate Planning Management of

Financial planning of production site or

company

Production planning, productivity planning at the product level

Maintenance, spare parts, logistics, with the management of the

Maintenance production or the

Quality control and management product area

Scheduling, time tracking of production, at the level of

Inventory: Product management (supervision)

Material, warehousing, logistics, procurement can still only be perceived by humans and this human component of automation must be further developed in order to enable the producers to control the production plant.

Regardless of the degree of automation, the operator on the lower levels is always chased by the following questions:

❖ Are the considered variable parameters sufficient to really control the production process?

❖ Does controlling one or some of the variables really mean controlling the overall process?

❖ Soft are the most important variables in the process?

❖ Which variables determine that the product has the desired specifications?

❖ What changes to the relevant variables are required for the product to meet the specifications again.

❖ Is there a model! of the entire production process?

These questions are brought together by the invention. The individual variables form the overall system. They are interconnected in a variety of ways. Therefore, the invention provides that the overall process is recorded in a model and that this model is stored in the process control and evaluated to control the process. For this purpose, it is first determined which physical and / or chemical parameters of the product and / or the production process characterize the product or the process so precisely that the quality of the product is within the specification and the quantity required is within the specified quantities. These parameters are called parameters. These parameters are influenced by adjustable and non-adjustable product parameters and / or process parameters. These so-called influencing variables must also be determined in detail.

It is then determined which effects on other product parameters or process parameters are caused by changes in the influencing variables become. The effect of changing the influencing variables on other influencing variables is also examined.

For this purpose, changes in the influencing variables are deliberately brought about in test series of the real process, or random changes are evaluated. In any case, a dependency model of the individual parameters and influencing variables is created, so that there is a network-like relationship between all influencing variables and parameters.

For this purpose, several selected physical parameters of the product or process are first measured and summarized in tables. The usability of these measured values for process control depends first of all on the fact that those parameters have really been recorded which are of significant importance for the quality and / or quantity of the product or the production. Likewise, outliers and other results must be eliminated, which have arisen due to coincidences, i.e. events outside of the planned normal course of the process.

But even then such a table with a large number of measured values and a large number of parameters is not meaningful and is unsuitable as a control means. Rather, an appropriate evaluation is required. Here, mathematical and / or statistical methods are helpful, which were developed especially for chemistry (chemometry). It is a matter of summarizing the measured values by means of special characteristic values (analysis values) in order to obtain information about the basic course of the measurement. For example, from the measurement results of a parameter Mean values, median values, standard deviations, variance, frequency distributions are formed.

FIG. 8 shows the measured values of an influencing variable X and a characteristic variable Y in a two-dimensional coordinate system. It was found in the test series that when different values for X are set, the measured values for Y are in a narrow range with a longitudinal extension. To reduce the two-dimensionality, the measured values can now be transformed into another coordinate system. For this purpose, an axis of the new coordinate system, namely the coordinate XX through the area in the direction of the longitudinal extent, that is z. B. placed on the median value of the measured values. This gives a new component of the process, which is referred to here as XX. This component XX is no longer dependent on the other component YY which is perpendicular thereto and also not — in three-dimensional space — on the vertical component ZZ. The components transferred into this coordinate system are no longer identical to the original parameters or their measured values, but are connected to them via the directional components of the new coordinates with respect to the original coordinates.

If the investigation of further influencing variables or parameters is now related to the new coordinate system XX-YY-ZZ, the other influencing variables or parameters are to be loaded (loaded) with the direction components of the new coordinates. In contrast, in this coordinate system the parameter Y does not occur in its dependence on the one-part size X, or only with its scatter in relation to the direction assumed for the new coordinate system, e.g. Median appearance.

Fig. Δ shows in a three-dimensional coordinate system such measurement values for a product, which represent the properties x, y and z of the product with two different setting variables. It can be seen that the majority of the measured values each lie in two clouds, each of which forms a narrow area with a clear longitudinal extension. The directional component of this longitudinal extension has a spatial position in the direction of each of the coordinates. The dimensionality is now reduced by transforming the measured values or analysis values (depending on what is shown in the diagram) to a new main axis of a new coordinate system, which lies in the longitudinal extent, for example on the median of one of the clouds. The zero point of the new coordinate system is preferably placed on the longitudinal extension of the other cloud, so that both clouds lie with their longitudinal extension in one of the coordinate planes. Here too, the original parameters are transformed into new components that are loaded with the directional components of the new coordinate system. Using this method, known as principal component analysis (see e.g. Christian Holm, Principal Components & Multidimensional Fit, in http://rkb.home.cern.ch), it is not only possible to reduce the dimensionality but also to discover the dependencies or non-dependencies between individual parameters and to quantify them to such an extent that a model of the process can be built up with them.

In this way, a mathematical model can be formed which is at least sufficiently correct for the normal operating range of the process.

However, a neural network can also be formed which, on the basis of the determined dependencies, allows, in the absence of a mathematical model, the generation of movements within this neural network itself.

The mathematical model or this neural network is continuously updated on the basis of experience, since the solutions determined by the network due to a change are constantly monitored using suitable measurements within the production process and the corrections are in turn stored within the mathematical model or within the neural network and saved and subsequently taken into account and evaluated accordingly.

Depending on the individual case, however, it may be preferred that this update is not carried out continuously, but is initiated in each individual case. It must be avoided that errors in the production process are included in the model process. If e.g. B. a part has become defective in the production process, e.g. B. a sensor delivers incorrect measured values, this error must not enter into the model process in the sense that a different kind of dependency is stored in the model process.

This is good not only for mathematically defined dependencies but also for dependencies that are stored in a neural network. For this reason, the planning and maintenance of the neural network and also the planning and maintenance of the mathematical dependency model represent an additional management task in the context of this invention, compare FIG. 5.

Such a model of mutual dependencies of parameters, their effects on a parameter and the feedback from the measurement The actual effects on the model determined are shown in FIG. 4. The parameters and influencing factors appearing in the individual levels are shown in their interdependency and linkage. It should be added that this is only a diagram from which no weighting of the parameters, no quantification and no weighting of the dependencies can be seen. However, it can be seen from the model that this model, when it is stored as a program and storage module in the control device of the production plant, does not appear to the operator, ie: is a "black box" for him. However, it allows the input of decisions and control variables in the levels provided - see level IV - with retroactive effect and action on all other levels.

With the constant adaptation of this model to the process actually driven, the values of the process variables and all other parameters of the process and the product, which are signaled continuously or not continuously by any measuring devices, are continuously recorded and the dependencies of these parameters laid down in the model process accordingly corrected the current values.

In order to achieve good results, the variables that are made the subject of the mathematical or neural network must effectively represent the process (parameters). Before planning the model (mathematical model or the neural network), it is therefore necessary that the technologists analyze the variables which they believe represent the process. This verifies the extent of their importance for the process. "Chemometry" means are available for this analysis. These are mathematical and statistical methods for the analysis and interpretation of complex multivariable systems, (see e.g. R.C. Graham, Data Analysis for the Chemical Science VCH Weinheim, 1993 and M. Otto, Analytische Chemie, VCH Weinheim 1995)

5 shows the levels 2 and 3 with the additional automation elements created by this invention, networking of the data flow, the information obtained therefrom and the resulting ones Control options shown again. These are in particular the following information complexes of level 2 obtained from the data of levels 0 and 1 and the following strategic decision elements resulting from this on level 3:

Level 2 Level 3 total statistical methods

Control diagrams quality management, non - linear techniques (neural networks) supplemented by the multivariable statistical methods

(Chemometry) data.

EDP monitoring with server - diagnosis / control

(Tracking) / and coincidence internal quality control / management planning and maintenance of the

(Chemometry) neural network

Samples management

Real quality control

With these means, it is possible not only to obtain statistical values and trends from the process data determined by measurement, but also to obtain explanations about the course of the process and its internal relationships, to determine which are the most important parameters and how they are related To determine the course and trend of the variables that are not continuously signaled and to generate ongoing control sheets. Above all, however, it is no longer only the course of individual variables that is shown, but the entire process in a matrix. This matrix contains the entire previously optimized dependency structure of the mainly influential process variables and other influencing variables. If this matrix is saved, it serves as the basis for controlling the process as a whole and the end product obtained from it in a control loop and for automatically optimizing it according to predefined rules and at specified times by constantly adding and updating the dependency structure of the influencing variables laid down in the matrix , In summary, the invention with its individual elements is shown again schematically in FIG. First of all, a series of tests on the real process and the product being produced must be carried out using a suitable selection of sensors to determine a grid of the production parameters, ie the influence sizes and the parameters. From these production parameters, those are selected that have the greatest influence on the quality of the product and / or the process. To reduce the dimensionality, if necessary, the production parameters or a group of production parameters or a plurality of groups of production parameters are each transformed into a different coordinate system in which at least one dependency ratio has been eliminated. This creates a network of dependencies of the selected parameters. This network of dependencies is specified as a model for a process computer. This process computer makes it possible to make a prediction of the result, in particular the quality of the process or product, for each planned input value on the basis of the fed-in parameters and dependency. In particular, it becomes possible to enter the same input values in the real process and in the parallel virtual process (model) while the real production process is running. The prediction provided by the model takes into account a large number of influencing variables and parameters in the previously entered complexity and operating ranges. The invention thus represents a departure from the previously customary method in which the individual measured variables (influencing variables or parameters) are isolated and incorporated into the network of action and counteraction of all parameters regardless of the embedding of this parameter.

This prediction by the model allows the influencing variables that were set for the desired prediction result in the model to be given to the real process. The results of the real process now supplied by the sensors can be compared with the results with the prediction of the model continuously or at time intervals. This in particular makes it possible to set up a control loop for the real production process and Adjust the input (input data) so that the actual result and the prediction result match as closely as possible.

In addition, the comparison of the prediction result and the actual result can also give reason to check the plausibility of the comparison result. For this purpose, the model contains a module that specifies the limits within which a deviation between the model result and the real result can be expected. If the defined area of plausibility is left, an intervention in the real process is necessary in order to remedy any errors. A violation of the plausibility exists in particular if the comparison result has a trend over time, i. H. grows continuously, even though the input (input data) remains constant.

However, this block can also define the influence that input changes may have. If certain values that have to be determined beforehand are exceeded here, in addition to checking the real process, a check of the model is also necessary in order to determine whether all influencing variables and parameters have been taken into account in a weighting that corresponds to reality.

It can be seen that here the management, i.e. Company management and plant management receive a comprehensive tool to make all important decisions competently and quickly in all stages of the production of a specific product, from product planning to production control! hold true.

Claims

Expectations 1 Method for carrying out an automated production process, in which production parameters are measured and the production process is controlled as a function of the measured values by a process computer, the production parameters being the parameters which are a measure of the quality and / or quantity of the product or of the production process, and the influencing variables which influence the parameters obtainable in the production process include, characterized in that a virtual model of the production process (dependency model) is built and operated in the process computer, in soft model only a limited selection of preferably less than 4 (four) production parameters with their real mutual dependencies previously determined on the basis of the production process is stored in such a way that the input of influencing variables into the virtual model result in the prediction of those parameters which are associated with di These influencing variables can be achieved as target values of the real process. 2 The method according to claim 1, characterized in that the influencing variables and / or parameters of the virtual model with measured influencing variables and / or parameters of the real production process (measured data compared and the deviations signaled and preferably signaled in classes and assigned to the products of the real production process 3 The method according to claim 1 or 2, characterized in that when comparing the influencing variables or parameters of the virtual model and measured influencing variables and / or parameters of the real production process (measurement data), it is determined whether the detected deviations in terms of their size and / or in terms of their temporal course lies within a predetermined confidence interval, which is specific for the respective influencing variables or parameters, and that a warning signal is issued when the trust area is left. Method according to one of claims 1 to 3, characterized in that the influencing variables and / or parameters of the virtual model with measured influencing variables and / or parameters of the real production process (measured data compared and the real production process as a function of the determined deviations between the model and the real process is controlled. Method according to one of claims 1 to 4, characterized in that in the real process those influencing variables are specified and set as sol variables which lead to the desired parameters in the model due to the stored interdependencies. Method according to one of claims 1 to 5, characterized in that the selection of the production parameters of the virtual process takes place on the basis of test series of the real process in which the mutual dependencies of the production parameters of the real process are evaluated and weighted. A method according to claim 6, characterized in that in the test series > it is determined which physical or chemical parameters of the Characterize the desired quality and / or quantity of the product and / or production process with sufficient accuracy (parameters); > it is determined which product parameters or process parameters, which can be set and which are not adjustable, but which can be changed, have an influence on the parameters (influence parameters); > by evaluating deliberate or accidental changes in the influencing variables with regard to their effect on predetermined parameters of the product (product parameters) or parameters of the process (process parameters or other influencing variables, the interdependency between the production parameters, in particular between the parameters and the influencing variables, is determined, > those maximum three production parameters are selected for the virtual process that best describe the virtual process in the desired range of the selected parameters due to the determined dependencies. Method according to one of Claims 1 to 7, characterized in that the analysis values obtained from the existing measurement values are those obtained using mathematical and / or statistical methods of chemometry, in particular mean value, meridian value, standard deviation, variance. Method according to one of Claims 1 to 8, characterized in that, in order to determine the interdependency between the influencing variables and the effect of the influencing variables on the parameters, the measurement values obtained are analyzed and interpreted using mathematical and statistical methods and that the results of the analysis (analysis values) a dependency model is developed and saved. Method according to one of claims 1 to 9, characterized in that the dependency model is formulated mathematically, preferably linearly. Method according to one of claims 1 to 9, characterized in that the dependency model as a neural network (semantic network, neural network) is designed. Method according to one of claims 1 to 11 Characteristic: of more than three measured adjustable influencing variables, those (two or three) influencing variables or analysis values are selected for the model, the measurement values or analysis values of which occupy an area with the greatest linear extension, and which accordingly are of greatest importance with regard to their interdependency and / or their impact on the parameters. Method according to one of Claims 1 to 12, characterized in that the number of measured production parameters is reduced to the number of model parameters entering the model by two measured, mutually dependent production parameters (parameters which are a measure of the quality and / or quantity of the Product or the production process, or influencing variables that influence the parameters that can be achieved in the production process) are transformed into a coordinate system, one axis of which lies in the direction of the greatest linear extension, in particular on the mean value or the median value, of the area in which the measured values are located or analysis value, so that the values of this axis serve as substitute parameters for the pair of production parameters, if necessary this process is also repeated with other pairs of dependent production parameters and / or dependent production parameters and substitute parameters, and that the remaining replacement parameters are included in the model as model parameters. Method according to one of claims 1 to 13, characterized in that The model is no more than three-dimensional, the model being created by transforming the measured values or the analysis values obtained therefrom into a two- or three-dimensional coordinate system, the main axes of which in each case in the direction of the greatest linear extent, in particular on the mean or the median, of those (two or three) ranges in which the measured values or analysis values of the selected two or three influencing variables lie Method according to one of claims 1 to 14, characterized in that in the model, the mutual dependencies of the influencing variables with one another and / or the mutual dependencies between the influencing variables and the parameters are corrected, supplemented or otherwise by evaluating current or temporary measurement data of the real processes Such updates will preferably be continuously updated automatically.