EP0813701B1 - Control system for e.g. primary-industry or manufacturing-industry facilities - Google Patents

Abstract

A method for controlling a primary industry plant of the processing industry, for example, in a steel plant or a rolling mill in order to, for instance, produce strips of steel or non-ferrous metals. The control method is designed in terms of computer engineering building on inputted advance knowledge, such that the method can automatically recognize the state of the installation and details of a manufacturing process taking place in the installation, for example in a continuous casting process for strips, and is able to give desired values and setpoints appropriate for the situation to achieve a reliable and successful production.

Description

The invention relates to a control system for a plant Basic material or the processing industry or the like, e.g. For a metallurgical plant, for example for the production of strips made of steel or non-ferrous metals, the control system using computer technology, building on previous knowledge, the state of the facility and details of one that is running in the facility Manufacturing process, e.g. a continuous casting process for tapes, automatically recognizing and for achieving safe production success according to the situation giving, is trained.

For industrial plants for the production or processing of There has always been a need for goods or energy Control system that provides an optimal and, in particular, inexpensive, Management of the process carried out in the plant enables. This need has so far been less than ideal through facilities of conventional control technology so far taken into account as possible. Especially in production processes, which cause major control problems, however, the necessary regulatory effort increases enormous, without the result really satisfactory is. For example, DE 43 01 130 Al a process and a unit for controlling a rolling stand, i.e. a single component. An optimization of the process flow This is at most based on the individual component under consideration.

For strip casting systems for metal, the operation of which is particularly large brings technical problems with them and is therefore exemplary to be dealt with further, it is already known with interconnected individual controllers or control loops to work. Examples show the EP 0 138 059 A1 and EP 0 228 038 and the essay "Development of twin-drum strip caster for stainless steel" by K. Yanagi et al., Metec Conference, June 1994, Mitsubishi Heavy Industries, Ltd./Nippon Steel Corp. The well-known Regulations that work suboptimally, although partially are already equipped with regulators, the mathematical Using models leads to the production of tapes whose Dimensional accuracy and quality are still relatively large fluctuations subject to. It is particularly disadvantageous that the systems, who work with the known controllers and control loops, need fast, preferably hydraulic, actuators, which are very expensive.

To at least partially address the above disadvantages avoid, it is known to use expert systems. Expert systems as so-called intelligent systems, with whom the quality of the manufactured products also in relation on quality features that are difficult to control are also to be improved for plants in the raw materials industry known, e.g. from the article "Process optimization for maximum availability in continuous casting ", published in the journal "Metallurgical Plant and Technologie International 5/1994 ". Such expert systems, that can certainly improve the production success however not the principal weaknesses of the conventional scheme. These will be especially then visible when it comes to processes that (such as in the raw materials industry) due to the lack of more suitable ones Sensors, for example inside high-temperature processes, do not can be regulated directly.

It is specially designed to control the strip casting of steel indirect regulation from EP 0 411 962 A2 known, with a family of curves of admissible input variables as Plant management base to work. The family of curves gives that Course of input variable constellations recognized as safe again. Such an approach, based on expert knowledge implemented in the system management via setpoint specifications is required in the event of changes in quality or requirements complex system behavior tests to determine new ones Leadership curves. Beyond that, there is only one working in a distance Distance from the process optimum possible.

It is an object of the invention, especially for conventional difficult to control production processes, e.g. for strip casting of metal strips to indicate a guidance system with which Better production success in low-cost systems can be achieved.

The task is actually intelligently trained Control system solved, based on the previous knowledge entered, automatically appropriate instructions for a situation safe and as good (optimal) process management as possible. It is therefore a fully trained technical Intelligence that, surprisingly, is already with today available computing resources also for Process control systems of large plants can be realized.

In an embodiment of the invention it is provided that the invention Control system the appropriate instructions is computationally optimizing step by step. This will further increase intelligent behavior achieved that lead to a quality of litigation leads that by human operators or not at least not in the short, computationally achievable Time is achievable.

In a further embodiment of the invention it is provided that the prior knowledge entered, that is, the one predefined by humans Process knowledge, preferably automatic, ongoing through on Process during production internally, e.g. in different operating points, knowledge gained improved and this self-generated process knowledge into one, in particular constantly updated, data storage as new Previous knowledge is adopted. So one becomes very advantageous constantly improving basis for further adaptation or Process optimization created. The knowledge gained is not only limited to more precise parameters etc. but also particularly closes the principles of algorithms etc. used.

In particular for achieving a safe production success, which is the basis of customer trust in one forms such a system, it is provided that the control system has a basic functional system for the system components, which the instructions from the computational, e.g. from a process model, preferably an overall process model, gained knowledge, safely implemented in the plant management. By connecting a safe basic function system, preferably as one of the system components each for yourself or in summary, safe to work Basic automation system is designed with an appropriate situation adapted, static process model results an execution related to the security of litigation the conventional training of a control system at least equivalent, in terms of cost / benefit ratio and the safe process result, but superior is.

It is particularly advantageous that the situation-appropriate Instructions, e.g. in the form of setting values, the system components directly in the form of control values, for example of positions or in particular indirectly, e.g. via controller setpoints, for speeds, for example. The Instructions are particularly advantageous directly from the Process model sizes determined. This happens for Time-critical setpoints are advantageous on-line, otherwise off-line. This results in a particularly favorable response from the system to changed process conditions under advantageously possible Saving of setpoint calculators.

This is advantageous for increasing operational safety Basic automation system as an autonomous, a safe one Condition of the system or system components and the process status guaranteeing subsystem, e.g. as a hazard state relapse system trained that, if necessary, instead of computationally generated instructions, especially as safely recognized operating values stored in the data memory can fall back. So that's a safe, if sub-optimal Working the system even in the event of a failure or Malfunction of the intelligent computer part occurs possible.

The basic function system advantageously also has start and Start-up routines that are entered manually or automatically can be, as well as suboptimal normal operating routines, in to those instructions, otherwise determined by calculation can be replaced by constant, safe guidelines. A is such a configuration of the basic function system especially for the commissioning phases, for operation with erratic change of requirements etc. advantageous. To when also suboptimal, function of the intelligent computer part do not always need all model parts in specific adapted form are available. It is also beneficial a company with a partially elaborated and / or adapted overall process model possible.

The process model itself is in the form of an overall process model modular structure and describes the behavior between the process input variables and the manipulated variables and the process output variables, e.g. Quality parameters of the product created. The modularity allows one particularly advantageous design and processing of Overall process model, as it is made up of individual, clearly visible, Part models can be assumed. The process model is based advantageous as far as possible on mathematical forms of description. Where such mathematical forms of description are not possible, for example, is formulated on linguistically Model parts used, e.g. through fuzzy systems, Neuro-fuzzy systems, expert systems or similar be realized can. For, e.g. completely new, system components for which no modeling based on mathematical-physical, chemical or metallurgical basics or similar or due to of linguistically describable process knowledge is possible, self-learning systems, e.g. neural networks used. This results in indifference for all production plants how big they are designed or how they are designed, the possibility to create an overall process model.

It is of course also possible to use the production process Parts for which inexpensive conventional solutions for Are available to operate conventionally. Then it will otherwise necessary model module taking into account the effect of the conventional part used replaced. This procedure may be available in the reel area of a rolling mill.

The process model becomes advantageous due to the on the plant collected process data, which is archived in a process database are continuously adapted to the process and further improved, this advantageously using adaptive methods, Learning methods, e.g. through a backpropagation learning process or a selection process for different sub-models, such as neural networks or parts thereof. So results a largely self-taught model, the can be adapted or improved on-line or off-line.

In an advantageous embodiment it is provided that the adjustable Process variables by the optimizer on the process model are optimized in such a way that the model output variables, which are in particular quality parameters of the product, as well as possible with given, e.g. the target, Values match. Through an offline processing of the The high computational effort of such processes becomes optimization manageable at low cost. The offline optimization can both on a separate processing unit parallel to the Model adaptation, as well as in breaks, e.g. at the weekend or in the event of repair downtimes on the computer, e.g. in the Operation of the reference variables of the basic functional system issues, take place.

The optimization is advantageously carried out using known optimization methods, especially about genetic algorithms. The The optimization process is selected according to the situation and problem-dependent. It can be done by a specification, e.g. based on an analysis of the course of the process, or by computational Selection from a collection of optimization methods respectively. A simple "trial and error" procedure can be used for this applied, but it is recommended to reduce it the computing effort, the "trial and error" procedure through convergence criteria, methods of pattern recognition in Support the error acceptance process etc.

The respective start values for an optimization are advantageous on the basis of the archived in a process data memory suboptimal operating data determined. So reduced the optimization effort because the optimization calculation starts with pre-optimized values if they are certain recognized intermediate values are used as start values.

The overall system is improved at least in three stages. The lowest level is the continuous improvement of the existing process knowledge, stored in the data memory, e.g. in Form of sub-optimal, safe operating points that automatically continuously brought to a better adapted level of knowledge , and from that again it is assumed becomes.

The second stage essentially consists of the model adaptation, that adapts the model behavior to the process behavior as well as possible.

The third stage is a continuous improvement of the situation-specific instructions from the process optimizer, e.g. about evolutionary strategies, genetic algorithms etc. These strategies require a lot of computing time and preferably run off-line.

The system improvement is also advantageous through external simulation calculations, model tests, possibly also through tests on the production plant with new aids etc., continuously supported.

The control system according to the invention is exemplified below using a steel strip caster. Here there are further, also inventive details and Advantages from the drawing and the description of the drawing as well as from the subclaims.

FIG 1

eine schematisierte Darstellung der Bandgießanlage mit Meßdatenerfassung und Stellgrößenausgabe,

FIG 2

die Struktur des "intelligenten" Teils des Leitsystems mit der Sollwert-Vorgabebildung,

FIG 3

Einzelheiten des Prozeßoptimierers,

FIG 4

Einzelheiten des Adaptionsvorgangs,

FIG 5

wesentliche Bestandteile des Prozeßmodells und ihre Grob-Verknüpfungsstruktur,

FIG 6

erfindungswesentliche Teile des Datenspeichers und

FIG 7

ein Komponenten-Schema der Basisautomatisierung.

In detail show:

FIG. 1

a schematic representation of the strip caster with measurement data acquisition and control variable output,

FIG 2

the structure of the "intelligent" part of the control system with the setpoint specification,

FIG 3

Details of the process optimizer,

FIG 4

Details of the adaptation process,

FIG 5

essential components of the process model and their rough link structure,

FIG 6

parts of the data memory essential to the invention and

FIG 7

a component diagram of basic automation.

In FIG. 1, 1 denotes the casting rolls of a two-roll casting device, the material between the casting rolls 1, liquid steel, from the ladle 4 over the tundish 5 and a dip tube 6 is entered and solidified into a band 3, that in one, through the circles 2 with movement arrows symbolized, rolling mill can be further deformed. The downstream rolling mill can also be easily moved by conveyor rollers, a reel or similar be replaced when rolling out should not be done immediately after pouring. The design the entire system is made according to requirements. Also training the, the pouring device downstream, system as a hot-cold rolling mill is possible and recommended at very high casting speeds because then the cold rolling section of the plant is sufficiently utilized can be.

Between the casting rolls and the downstream equipment the casting and rolling device preferably also has one only symbolically represented electrodynamic system 8.9 and an induction heating system 10. The electrodynamic System part 8 advantageously serves to relieve weight, the cast, here still very soft and therefore at risk of constriction, Volume 3 and the electrodynamic system part 9 of guiding the tape 3 during the induction heating system 10 compliance with a predetermined temperature profile over the bandwidth if, e.g. a direct one Post-forming in a rolling mill. This is particularly advantageous for crack-sensitive steels. The The cast strip 3 is checked for cracks by a camera 73, which can be used to advantage that the crack pattern in the scale is influenced by cracks in the base material becomes. The formation of a measured variable is advantageous through a neuro-fuzzy system.

Because the surface temperature of the casting rolls to avoid of temperature change stresses essentially constant should be, they are through an IR heating system 7, an induction heating system or similar even in that, not with liquid steel area in contact, kept at working temperature. These and other individual components of the, only roughly schematic drawn, casting and rolling device are e.g. via temperature controller, Flow adjuster, speed controller etc. in Framework of basic automation via a manipulated variable output 12 set directly or regulated. Actual data of the actuators, the controller etc. are in the measurement data acquisition 11 for the data storage and the model input as well as in not summarized as shown for the basic automation and processed. Through data transfers I, II and VI, which are symbolized by arrows is the casting and rolling device, in which the formed on the two casting rolls 1 Solidification shells of the steel not only combined, but can also be rolled pre-formed with the intelligent Part of the control system connected.

2 shows the structure of the intelligent part of the control system. This essentially consists of the parts process optimizer 15, model 20, model adaptation 16 and data storage 17. These parts of the control system interact in such a way that as good as possible via the setpoint output 13, Instructions appropriate to the situation via the data line V for Litigation can be made available. This Instructions are then converted into setpoints for basic automation implemented. The following is the task and the function of the individual parts is described.

von den Stellgrößen u_i , mit denen der Prozeß beeinflußt werden kann, und von den nichtbeeinflußbaren Prozeßgrößen v_i , wie z.B. der Kühlwassertemperatur, nach. Die Modellausgangsgrößen sind dabei, wie schon erwähnt, typische Qualitätsparameter des Produktes. Die Modellbeschreibung

erfaßt das Prozeßverhalten im allgemeinen nicht exakt, weshalb y_i und mehr oder weniger voneinander abweichen. Übertragen werden die Stellgrößen u_i und die nichtbeeinflußbaren Stellgrößen v_i über die Datenleitungen I und II.The model 20 forms the static process behavior y i = f i ( u l , ..., u i , ..., v l , ..., v i , ...), ie the dependency of the n model output variables

from the manipulated variables u _i with which the process can be influenced and from the non-influenceable process variables v _i , such as the cooling water temperature. As already mentioned, the model output variables are typical quality parameters of the product. The model description

generally does not exactly grasp the process behavior, which is why y _i and deviate more or less from each other. The manipulated variables u _i and the non-influenceable manipulated variables v _{i are} transmitted via data lines I and II.

The model adaptation 16 has the task of improving the model, thus the model behavior as well as the process behavior corresponds. This can - at least for model parts - done online by making these model parts based on continuously recorded process data can be adapted or updated.

minimiert in Abhängigkeit von den Modellparametern oder der Modellstruktur. D.h. man variiert die Modellparameter bzw. die -struktur so, daß ε möglichst klein wird.For other model parts, the adaptation can also be carried out off-line at certain times. This is done on the basis of a number m of process states representing the process ( u ^k _i , v ^k _i , y ^k _i ), which are stored in the data memory 17. The index k quantifies the respective process status. With this type of adaptation, the model error

minimized depending on the model parameters or the model structure. This means that the model parameters or structure are varied so that ε becomes as small as possible.

The process optimizer has the task of using an optimization method and the process model to find manipulated variables u _i which lead to the best possible process behavior. The process optimizer works off-line at specific times, for example, which can be predetermined manually, as follows:

bestimmt.First, the manipulated variables v _i that cannot be influenced, for which the optimization is to take place — for example the current ones — are kept constant and are fed to the model via the data line II. The process optimizer is then connected to the model by means of switch 18. It gives control values u _i to the model. The initial values are based on the model certainly. These are compared with target output values y _{target, i} , and it becomes the error

certainly.

The error E should be minimized. For this purpose, the process optimizer varies the manipulated variables u _i in an iterative loop, which contains the calculation of y _i and E as well as the new selection of u _i , until the error cannot be reduced further or this optimization is terminated. Genetic algorithms, hill climbing methods, etc. can be used as optimization methods.

The optimal manipulated variables u _{opt, i obtained in this way} , which are the result of the above-mentioned minimization, are then transferred to the basic function system via the setpoint specification and the data line V as setpoints.

The main task of the data memory is to archive representative process states ( u _i , v _i , y _i ). In doing so, he replaces old process data again and again with newly determined one, in order to enable a current, albeit selective, process description based on this data. The data memory then supplies the model adaptation, as described above. On the other hand, it also supplies start values u _i for the process optimizer. The start values are selected, for example, in such a way that the output values y _i belonging to these start values correspond as well as possible to the target values y _{target, i} .

The preferably off-line loop: model 20 and Process optimizer 15, e.g. genetic algorithms for e.g. evolutionary, model improvement, works preferably off-line because of the complexity of a plant control model with its many possible Develop the computing time of an evolutionary optimization process becomes comparatively long. Even with good ones Optimization strategies that e.g. based on an analysis of the probable model behavior are selected many optimization processes until a clear one is reached Calculate model improvement.

The creation of a model structure to be used according to the invention and an essential sub-model is e.g. by doing "Automation Of A Laboratory Plant For Direct Casting Of Thin Steel Strips "by S. Bernhard, M. Enning and H. Rabe in "Control Eng. Practice", Vol.2, No.6, page 961-967, 1994, Elsevier Science Ltd. described. From this publication include also the basic structures of suitable basic automation systems and can be seen from start routines which the specialist can build up.

As a computer for process optimization and parameter adaptation are workstations, e.g. from the Sun company. For large control systems, it is advantageous to work in parallel Calculator used. This is especially true if the model can be divided into groups of model modules that are part dependent can be optimized from each other.

At comparison point 19, into which the setpoints, in the selected exemplary embodiment the setpoints for the strip thickness, the profile shape, the surface quality of the strip, etc., are incorporated, the results from the model calculation are continuously compared with the setpoint values. The difference can be minimized through optimization. Since the difference in technical processes cannot generally become zero, the optimization process must be meaningfully limited, that is, it must be terminated. FIG. 3 shows more precise details of the program structure with which the optimization is terminated and the new setpoint output is started.
In FIG. 3, 58 denotes an error function to be selected in each case, into which the detected errors (setpoint deviations) flow. 61 now examines whether the error function meets the termination criteria of the optimization. If this is the case, further optimized control and regulating variables are output. Before the termination criterion is reached, start values from the data memory continuously enter the start value specification 59, from which, in search steps in 60, not from the optimizer, but from the data memory, for example with the aid of fuzzy interpolation, control and regulating parameters for sub-optional process control are obtained. Switching takes place after reaching the predetermined quality factor, which is adapted to the respective control system knowledge. As already said above, the minimization, which can never be absolute, is stopped when the predetermined quality factor is reached.

The model will also be advantageous if it goes to the Process connected, i.e. Switch 1 is closed, too generates an alarm signal that is critical Operating states signaled. Such procedures are already known and can be found in the same way in conventional control systems.

In FIG 4, which the structure of a model adaptation by means of a Optimization algorithm shows data from the start value specification 61 in a search step unit 62 and are from passed there as model parameters to model 63. The Model 63 forms, together with data storage 64, a parameter improvement loop, the 65 in a known manner the compares formed and stored values. The comparison values are fed to the error function 67, their Passes values to the termination criteria unit 66. Are the If the termination criteria are met, the model will not continue improved and worked with the existing values. Otherwise the optimization with further search steps and the Intermediate values continued in the data storage.

In FIG 5, the essential sub-models of the overall process model of the embodiment, 46 denotes that Input model in which the external influences, such as the influences summarized from the quality of the material used are. The quality of the steel used results e.g. of the Liquidus value, the solidus value, as well as others, the casting behavior characteristic sizes. 47 denotes the tundish model, into e.g. the steel volume of the tundish, the dip tube position or the like, the stopper position and the steel discharge temperature come in. The input models 46 and 47 are summarized in sub-model 56, which the status of the fed Material. Such partial models can advantageously parallel to other sub-models, such as the casting area model, the rolling range model or similar be optimized.

The input model 48 contains the influences that the solidification influence, e.g. the casting roll cooling, the infrared heating etc., the input model 49 contains the values that necessary for the heat balance, such as the steel casting roll temperature difference, the influence of lubricant as a function the amount of lubricant, the rate of crystal formation the respective steel grade and e.g. the roll surface condition. The input model 50 contains e.g. the influences of Casting level characteristics, so the level of the casting level Slag layer thickness and the radiation coefficient. The Input models 48, 49 and 50 are part of a model 54, the reflects the status of the casting area, summarized. This Model area summary is general for production areas advantageous as it is the overall model optimization simplified and improved. The sub-models are among themselves partly still dependent on each other, for example to a considerable extent the input models 49 (input model heat balance) and 50 (Initial model of the mold level characteristic). Secondary dependencies are not shown for simplification.

The sub-model 51 contains all influences on the solidification front, i.e. to the area where the on the two chill rolls solidified metal shells meet. Essentially these influences are the forming work done by the casting rolls is achieved, the vibration width of the casting rolls or the emerging tape, the side gap sealing influences and the level of effort of the overall system, this is e.g. a fuzzy model. The sub-model 52 gives the exit values again, e.g. the quality of the tape that Outlet temperature and distribution, but also the tendency to stick and the condition of the scale formed. In the sub-model 52 also goes the input model 53 and the input model 74 a, which relates to the temperature profile across Tape and refer to the surface condition of the tape. For the particularly advantageous case that it is a The strip casting and rolling mill also works Rolling mill part models 54 in this special process model a, since the product training after leaving the Roll stands is the decisive criterion.

The sub-models are combined to form the product training model 57, which is the thickness profile of the band formed, the strip thickness, a possible error pattern, the grain structure of the tape, the surface structure etc. summarized. The surface structure and especially the grain structure of the tape can only be determined with a considerable time delay. It is therefore advantageous to work with partial models here on the basis of neural networks for qualitative and quantitative determination of influencing variables.

From the above illustration, the special one emerges Advantage resulting from the training of the model in module form results, in particular, as the parts of a complex overall process model can be processed in parallel. This is special advantageous for the commissioning period of a plant, in which the input and partial models reflect the actual circumstances adjusted, linked together, etc. must be.

6 finally shows the part that is essential according to the invention the data storage structure. 68 denotes the process data archive , 69 the model parameter storage part, 70 the part with the start values for the optimizer and 71 the memory part for the safe operating points. In 68 the respective Model training saved.

The basic automation, with its regulations, controls, Interlocks etc., an indispensable part of the Control system, because it the safe functioning of the Attachment even if the model part of the Guaranteed control system according to the invention, must Perform a variety of functions.

The individual functions are not, finally, due to the symbolized individual "black box" in FIG 7. Here means 21 in the embodiment, the mass flow control over the Single speed controller, 22 the regulation of the tundish heating, 23 the mold level control, 24 the tundish outflow control and 25 the heating power of the infrared or similar. Screen 7 for the Maintaining the operating temperature of the casting rolls. 26 means the regulation of the addition of lubricant, e.g. in shape of loose mold powder or of one applied to the casting rollers Casting powder paste, 27 the cooling water quantity control, 28 if necessary the roller oscillation control, 29 the electrical Drive control and 30 the roll gap setting. 31 means the roller speed control and 32 if necessary the control of the Roller torque, 33 the setting of the cleaning system, consisting for example of a brush and a scraper for the casting rolls and 34 the regulation of the electrodynamic System for balancing the belt weight and the regulation the vibration width of the cast strip. 36 means that Regulation of the individual parts of an electrodynamic system for side gap sealing and 37 the regulation of the heating for the side walls of the space between the casting rolls. 38 means the temperature profile control of the induction heating system 10. 39 as well as indicated other control units refer to regulations of the downstream deformation units, e.g. Rolling stands, the train between them Roll stands etc. On the above actuators, controllers etc. acts the time control 45, which outputs the manipulated variables etc. coordinated in time. In block 40 are the example Auxiliary controls and the interlocks summarized so mean e.g. 41 the automatic start-up, 42 the automatic switch-off, 43 and 44 interlocks, e.g. prevent Liquid steel can flow before the pair of casting rolls and the Deformation rollers are workable, etc. In addition, they are further systems, not shown in the schematic, for the band edge separation required, e.g. by Lasers, for influencing scale formation, e.g. by Silicating, roller lubrication etc. available. In the Basic automation in which the measurement data I and the setpoint specifications V are received, the manipulated variables VI are generated, over which the system is managed.

The characteristic of self-optimizing and knowledge-based further developing control system, using the example of the Cast rolling process shown are explained in more detail below:

The casting and rolling process consists of a number of sub-processes, their training and influences are decisive for the End product. Can be influenced and optimized according to the invention are the properties of the end product, e.g. its Thickness, its thickness profile and its surface formation, through a number of adjustable process variables, e.g. the Casting roll gap, the casting roll profile, the casting level etc., which in turn shows the location of the union zone on the Casting rollers affect separated, solidified metal shells. For regulation and optimization is advantageous created an overall process model according to the invention, which the Process behavior describes. Based on this process model can be the influencing factors with which the process influenced, step by step according to the process conditions be adapted and optimized. That through this optimization certain situation-specific instructions then lead to an improvement in the process. Total result despite the relatively time-consuming process of creating, ( can also be used with other systems with less effort), Software significant cost advantages because of the System with much simpler mechanical components, fewer controllers etc. can work than the known ones Investments. The sensor technology is also much easier, since only the process output variables must be continuously recorded precisely.

The intelligent, self-improving, Part of the control system consisting of three essential elements: The process model, the model adaptation and the process optimizer. The process model consists of subsystems (Modules) together, which vary depending on the process knowledge Will be type. With knowledge of the physical Connections can be classic, physical-mathematical Models are created. On the other hand, you only have experience or estimates, so are fuzzy or neuro-fuzzy systems related. In case you have little or nothing about knows the process behavior, such as cracking and At least in the beginning, surface training is set to neural Networks for process formation. Describes overall the model the relationship between the process variables, as in selected example of the mold level, the condition values and the quality of the potted material, the setting values the casting rolls etc. and the quality parameters of the Tape, e.g. the thickness, the profile and the surface formation.

Since the model may substantial, percentage based on uncertain knowledge, it is not exact. The The model must therefore be adapted and changed based on the process data obtained etc. This happens advantageously on the one hand using the well-known model adaptation based on data from past Process states set up. Based on this data the model parameters or similar so that the model behavior corresponds as closely as possible to that of the process. Also be modifying the models themselves, e.g. through genetic Algorithms, a combinatorial evolution etc. Corresponding Optimization strategies are known, e.g. from Ulrich Hoffmann, Hanns Hofmann "Introduction to Optimization", Verlag Chemie GmbH, 1971 Weinheim / Bergstrasse; H.P. sulfur "Numerical optimization of computer models using the Evolution Strategy, Basel, Stuttgart: Birkhäuser 1977; Eberhard Schöneburg "Genetic Algorithms and Evolutionary Strategies, Bonn, Paris, Reading, Mass, Addison-Wesley, 1994; Jochen Heistermann "Genetic Algorithms: Theory and practice of evolutionary optimization, Stuttgart, Leipzig, Teubner, 1994 (Teubner texts on computer science; Vol 9)

By the control system according to the invention with that described above procedure according to the invention becomes the previous one Leave the structure of a control system. About basic automation, which essentially concerns the process level (Level I), there is only one, intelligent Control system to which the production setpoints are specified and from it all default sizes (positioning commands) generated (Level II). In intelligent self-optimization it ensures based on the process result already achieved for ever better process results. Individual feedback loops can be omitted. Just for checking the Process results are quality control sensors necessary. The control system according to the invention therefore only has two more essential levels, of which the intelligent No visualization except for programming requirement. The elements of the basic automation can be used as a control be visualized in a known manner.

Claims (20)

1. Control system for a primary-industry or processing-industry plant, for example for a metallurgical-engineering plant, for producing strips of steel or non-ferrous metals for instance, wherein the control system is designed by computer technology, building on prior knowledge that is input, automatically identifying the state of the plant and details of a manufacturing process taking place in the plant, for example a continuous casting process for strips (3), and giving setpoint inputs (V) related to the situation in order to achieve production success that is reliable and, if applicable, as high as possible, characterised in that the control system simulates the state of the plant and the individual components of the plant for the purposes of optimization continuously with the aid of a process model (20, 63) that has partial models (48-47) and is set up in a modular fashion and which describes the behaviour between the process input variables and also manipulated variables and the process output variables, for example characteristic quality values of the product produced.
2. Control system according to claim 1, characterised in that it is designed so as to optimize the setpoint inputs (V) related to the situation, preferably automatically effecting the optimization in given optimization routines.
3. Control system according to claim 1 or 2, characterised in that the prior knowledge that is input is improved, preferably automatically, continuously by means of knowledge obtained at the process model (20) during production internally by means of computing techniques, for example at different operating points, and this self-generated process knowledge is taken over as new prior knowledge in a data store (17), in particular a data store that is constantly being updated.
4. Control system according to claim 1, 2 or 3, characterised in that the setpoint inputs (V) related to the situation, for example in the form of set values, are delivered to the plant components directly in the form of selection values, for instance of positions, or in particular indirectly, for example by way of controller setpoints, for instance for rotational speeds.
5. Control system according to claim 1, 2, 3 or 4, characterised in that it has a basic function system for the plant components that converts the setpoint inputs (V) from the knowledge obtained by means of computing techniques, for example from a process model (20), preferably a total process model, reliably into the plant control.
6. Control system according to claim 5, characterised in that the basic function system is designed as a basic automation system reliably rendering the plant components operational, either separately or together.
7. Control system according to claim 5 or 6, characterised in that the basic function system obtains its setpoint values directly from the intelligent portion of the control system that determines these values from the results of adaptation and/or optimization processes at the process model (20).
8. Control system according to claim 5, 6 or 7, characterised in that the basic automation system is designed as an autonomous subsystem (emergency-condition release system) that guarantees a reliable state of the plant and the process and which, instead of having recourse to the setpoint inputs (V) produced by means of computing techniques, in particular can have recourse to operating values which are identified as being reliable and stored in the data store.
9. Control system according to claim 5, 6, 7 or 8, characterised in that the basic function system has starting and run-up routines, which can be input manually or automatically, and also suboptimum normal operating routines, in which individual setpoint inputs (V), otherwise determined by means of computing techniques, can be replaced by constant setpoints.
10. Control system according to claim 1, characterised in that the process model has mathematical descriptive forms at least in part, in so far as it can be modelled on the basis of mathematical-physical, chemical or metallurgical or biological laws.
11. Control system according to claim 1 or 10, characterised in that for the plant components, for which process knowledge exists that can only be expressed linguistically, the process model has linguistically formulated model portions, which can be realized, for example, by fuzzy systems, neuro-fuzzy systems, expert systems or tabular compilations.
12. Control system according to claim 1, 10 or 11, characterised in that for the plant components, for which it is not possible to produce a model on the basis of mathematical-physical, chemical or metallurgical or biological fundamental principles or on the basis of process knowledge that can be described linguistically, the process model has self-learning systems, for example neural networks.
13. Control system according to claim 1, 10, 11 or 12, characterised in that the process model is continuously adapted to the process or corrected on the basis of process data that is collected at the plant and is filed in a process data base and in that this occurs by means of adaptive methods or learning methods, for example by means of a back-propagation learning method or a selection method for various partial models, for instance neural networks.
14. Control system according to claim 1, 10, 11, 12 or 13, characterised in that the adjustable process variables are optimized off-line by means of an optimizer at the process model in such a way that the model output variables, which in particular are characteristic quality values of the product, correspond, to the greatest possible extent, with given values, for example those to be striven for.
15. Control system according to claim 1, 10, 11, 12, 13 or 14, characterised in that the optimization is effected with a known optimization method, for example with a genetic algorithm, the Hooke-Jeeves method, a simulated annealing method or the like, and in that the respective optimization methods applied are specified as a function of the situation and the problem or are selected from a data file, for example as a function of the number of variables to be optimized and/or the formation of the minima to be expected.
16. Control system according to claim 15, characterised in that the criteria for abortion of the optimization methods, for example with neural networks, are determined in accordance with a method of pattern recognition or classical convergence criteria on the basis of the course of optimization.
17. Control system according to claim 12, 13, 14 15 or 16, characterised in that the starting values for optimization are determined on the basis of the suboptimum operating data that is filed in a process data store.
18. Control system according to claim 1, 10, 11, 12, 13, 14, 15, 16 or 17, characterised in that the optimization is effected off-line with the aid of the process model, wherein adjustable process variables, which have been determined in such a way that the characteristic values of the product produced, which characteristic values are simulated by the model, correspond, to the greatest possible extent, with the given desired values, are passed as setpoint values to the basic function system of the process and the process is adjusted by the latter in accordance with the setpoint values.
19. Control system according to claims 1, 10, 11, 12, 13, 14, 15, 16, 17 or 18, characterised in that in the case of a malfunction or the like of the model or the optimizer the setpoint values for the basic function system can be generated directly from the data of the process data base, interpolation being effected for the purpose of improving the setpoint values, in particular between the stored operating data.
20. Control system according to one of claims 1 or 10 to 18, characterised in that the model, for instance in the case of a rolling casting process for metal strips, takes into account in particular the restrictions of the manipulated variables, the actuator time response and, if applicable, the process dynamics, preferably in and before the region of the casting rollers, for example in relation to the position of the solidification-shell merging zone for the solidification shells deposited on the casting rollers.

Images