Title of Invention

METHOD FOR THE DESIGN OR CONTROL OF THE PROCESS SEQUENCE OF A PLANT IN THE BASIC MATERIALS INDUSTRY

Abstract

The invention relates to a method for the design or control of the process sequence of a plant in the basic materials industry, in particular a steel mill or rolling mill, decision variables about the process sequence being optimized by means of a mathematical optimization algorithm which optimizes the decision variables about the process sequence on the basis of a process model, the process model being distributed to two model levels, a higher-order model level (1) and a lower model level (2), the lower model level (2) having partial models ( 3, 4, 5, 6) which are linked by at least one model (7) at the higher-order model level.

Full Text

-1A- Description Method for the design or control of the process sequence of a plant in the basic materials industry The invention relates to a method for the control of the process sequence; of a plant in the basic materials industry, in particular a steel mill, the decision variables about the process sequence being optimized by means of a mathematical optimization algorithm which optimizes the settings of the process sequence on the basis of a process model. In order to control the process sequence of plant in the basic materials industry, it is known to optimize decision variables such as settings of the process sequence. It is furthermore known to carry out such an optimization by means of a simplified model of the process sequence. If the reactions of the process sequence, which are necessary for optimization, to settings of the process sequence are not all known, this method fails. The settings of the process sequence then have to be made heuristically. This applies, for example, to the calculation of pass plans in a rolling mill. If the modelling of the process sequence enables its optimization, then because of the necessary simplifications this optimization leads to suboptimal solutions for the settings of the process sequence. If heuristic methods for the settings of the process sequence have to be used, then the determined settings often lie a long way away from the optimal settings. The object of the invention is to specify a method and equipment by means of which the abovementioned disadvantages may be avoided, in particular with a tolerable economic outlay. In this case it is particularly desirable to specify a method and equipment which makes it possible to generate settings for a rolling mill in accordance with optimal pass plans. According to the invention, the object is - 2 - achieved by the distinguishing featuces of the invention. In this way, it is possible to make available to an optimization algorithm a particularly detailed model of the process sequence, so that the said optimization algorithm can determine optimal settings of the process sequence. In this case, at the lower model' level models can be used which do not allow specific optimization algorithms, such as gradient methods. Models of this type are, for example, systems of differential equations, neural networks, finite-element models, rule-based models or fuzzy models. In particular for processes, such as rolling mills, for example, which cannot be described with a tolerable economic outlay using equation systems, but for which for specific partial components, for example models in the form of neural networks, are available, the method according to the invention allows optimization to be carried out by means of an optimization algorithm, such as the gradient method. In this way it is possible to, optimize the decision variables of process sequences having a high complexity, such as for example in plant in the basic materials industry. In an advantageous refinement of the invention, already extant or known partial models, such as are employed, for example, in the process control, are used. This enables particularly cost-effective modelling of the process sequence. Further advantages and inventive details emerge from the following description of exemplary embodiments, with reference to the accompanying drawings and in conjunction with the subclaims. In detail: FIG 1 shows a two-level optimization, FIG 2 shows a target tree. FIG 1 shows a two-level optimization. In this case the process sequence is modelled at two modelling levels, a higher-order modelling level 1 and a lower modelling level 2. At the lower modelling level 2 there are partial models 3, 4, 5, '6, which model parts of the process to be optimized. At the higher-order modelling - 3 - level 1 there is a higher-order process model 7, which has model equations and/or model inequalities and which, together with the partial models 3, 4, 5, 6 of the lower modelling level, forms an overall picture of the process to be optimized. This two-level structure according to the invention for the optimization of a process sequence has been tried and tested, in particular for complex plant in which partial models, for example in the form of systems of differential equations, neural networks, finite-element models, rule-based models or fuzzy models, already exist. In the course of interlinking the models 3, 4, 5, 6, 7 of the two modelling levels 1, 2, variable quantities 10 are generated by the higher-order model 7, these quantities forming the input variables for the partial models 3, 4, 5, 6. Output variables of the partial models 3, 4, 5, 6 are in turn parameters 9 which enter into the higher-order process model 7 as input variables. Further input variables into the higher-order process model are presets 8. The object of the optimization is to find decision variables about the process sequence, such as for example settings of the process sequence. These decision variables form a subset of the variable quantities 10. Within the context of the optimization, firstly variable quantities 10, from which the partial models 3, 4, 5, 6 calculate parameters 9, are determined by the higher-order process model 7. On the basis of these parameters 9, in accordance with the selected optimization algorithm, new variable quantities 10 are formed. This cycle is repeated until a selected break-off criterion for the optimization is satisfied. In this case, two optimization operations proceed. Firstly, the model equations and the model inequalities of the higher-order process model 7 are optimized by means of target functions. The result of this optimization is variable quantities 10 which are optimal on the basis of given parameters 9. After this optimization has been completed, on the basis of these optimal variable quantities 10, parameters 9 are calculated by the partial models. On the - 4 - basis of these new parameters 9, new optimal variable quantities 10 are determined in the course of the inner optimization. On this basis of these new variable quantities 10, new parameters 9 are calculated in the partial models 3, 4, 5, 6. In the course of this outer optimization, iteration continues until the new parameters 9 differ by less than a tolerance value from the old parameters 9, or the new variable quantities 10 differ by less than a tolerance value from the old variable quantities 10. This method has been proven to be particularly suitable for determining pass plans, that is to say presettings for roll trains. In this case, methods of non-linear optimization are particularly suitable. For the pass-plan calculation, it is preferable if optimization criteria are formulated in the form of an under-determined system of equations, so that a solution space in the form of a compromise set results during the optimization. The formulation of the optimization criterion is in this case advantageously performed in the form of functions which are formulated as targets in a target tree, as shown in FIG 2. To this end, all the practical targets Zl, Z2, Z3, Za for a specific application are used. These targets can be refined further, so that a tree structure as in FIG 2 results. In this case, not all the criteria necessarily have the same branching depth. For the pass-plan calculation, it is possible, for example, for the following targets to result, with reference to FIG 2: Z optimal pass plan, Zl technical aspects, Z2 flexibility, Za economic aspects, Z12 control, Z13 engineering, Z121 product quality, Z1211 flatness, Z1212 structure, Z122 critical states, Z1221 maximum rolling force, Z1222 maximum motor power, Z1223 maximum rolling speed, Zal characteristic numbers, Za2 profit, Za21 proceeds, Za2 2 expenditure. As a result of the refining in the target tree, targets of a general type are described by more and more concrete targets which are described as leaves in the - 5 - last refining stage. The target tree which is produced in this way constitutes a hierarchical structure of targets, targets within one branch being complementary and targets of different branches generally being competitive. The target tree is advantageously defined to apply as generally as possible, so that in the case of a concrete application, only a partial structure of the target tree is used. For a concrete application, the individual optimization criteria are provided with attributes which define the manner of their use. In this case, the following attributes are applied: "not active" "target function" "secondary condition" or "assessment criterion". In this case, the attribute "not active" means that, in the concrete application, the target is estimated to be not relevant and thus is excluded from the further considerations. All the targets provided with the attribute "target function" are bases for the optimization, the result of which is a compromise set. This compromise set is reduced to a reduced compromise set using the targets provided with the "secondary condition" attribute. From this reduced compromise set, a ranking sequence for the decision variables about the process sequence is in turn determined by means of the targets provided with the "assessment criterion" attribute. This procedure, consisting of optimization, reduction of the compromise set and assessment of the reduced compromise set, is referred to as decision-making. Particularly suitable applications for the method according to the invention are represented by the calculation of presettings for the equipment of a roll train, analyses and elaboration of rolling strategies, planning processes during rolling operations, and supporting decision-making during the design of roll trains. 6 We Claim 1. Method for the design or control of the process sequence of a plant in the basic materials industry, in particular a steel mill or rolling mill, decision variables about the process sequence being optimized by means of a logic rule which optimizes the decision variabfes about the process sequence on the basis of a process model, characterized in that the process model is distributed to two model levels, a higher-order model level (1), and a lower model level (2), the lower model level (2) having partial models ( 3, 4, 5, 6) which are linked by at least one model (7) at the higher order model level (1). 2. Method as claimed in claim 1, wherein at the higher-order model level, variable quantities (10), of which the decision variables are a subset, are determined, parameters (9) being determined at the tower model level on the basis of the variable quantities (10), on the basis of which parameters the variable quantities (10) are in turn determined. 3. Method as claimed in claim 1 or 2, wherein the model (7) at the higher- order level (1) is an analytical model. 4. Method as claimed in claim 1, 2 or 3, wherein the partial models (3, 4, 5, 6) are systems of neural networks, fuzzy models or rule-based models, the partial models ( 3, 4,5, 6) advantageously being known or already extant models which are employed in the process control. 5. Method as claimed in claim 1, 2, 3, or 4, wherein the optimization is performed in the form of an optimization, for example in different variants of the known logic rules. 5. 7 6. Method as claimed in claim 1, wherein for the purpose of the outer optimization, the decision variables about the process sequence are iterated over both levels, that is to say over the higher-order model level (1) and the lower model level (8), until the parameters (9) determined at the lower model level (2) from the respectively preceding iteration step differ by less than predefined tolerance values from the parameters (9) determined in the current iteration step. 7. Method as claimed in one of the preceding claims, whereirvsome iteration steps of the outer optimization, in particular in the first iteration steps, the partial models (3, 4r 5,6) of the lower model level are linearized and, in the corresponding iteration steps, the linearized partial models are used instead of the original partial models (3,4, 5, 6). 8. Method as claimed in one of the preceding claims, wherein targets for decision making, of which the optimization is a part, are formulated in a hierarchial manner, i.e. targets within one branch being complementary and targets of different branches preferably competing, the targets becoming more concrete with increasing branching. 9. Method as claimed in one of claims 1 to 8, wherein a target is taken into account as a target function, as a secondary condition, as an assessment criterion or even not at all during the dectsion-making. 10.Method as claimed in claim 9, wherein for the purpose of decision-making a reduced compromise set is determined from the compromise set by means of the secondary conditions and a ranking sequence is determined, in particular point by point, from the reduced compromise set by means of the assessment criteria., 11,Method as claimed in one of the preceding claims, wherein the method determines optimal pass plans for a roll.train, with the appropriate settings of the roll train. 8 12.Method as claimed in claim 11, wherein the method determines for existing roil trains optimal pass plans or rolling strategies with the appropriate settings of the roil train. 13.Method as claimed in claim 10 or 11, wherein the optimal pass plans are used as the basis for organizing the rolling operations (inter alia, of the ingot store) within the context of level -3 automation. 14.Method as claimed in one of the preceding claims, wherein the models (7) of the higher order model level (1) are not implemented on the same computer as the partial models (3,4, 5, 6) at the lower model level (2). 15.Method as claimed in one of the preceding claims, wherein the partial models (3, 4, 5, 6) at the lower model level (2) are in part implemented on different computers. The invention relates to a method for the design or control of the process sequence of a plant in the basic materials industry, in particular a steel mill or rolling mill, decision variables about the process sequence being optimized by means of a mathematical optimization algorithm which optimizes the decision variables about the process sequence on the basis of a process model, the process model being distributed to two model levels, a higher-order model level (1) and a lower model level (2), the lower model level (2) having partial models ( 3, 4, 5, 6) which are linked by at least one model (7) at the higher-order model level.

Documents:

01566-cal-1997 abstract.pdf

01566-cal-1997 claims.pdf

01566-cal-1997 correspondence-1.1.pdf

01566-cal-1997 correspondence.pdf

01566-cal-1997 description (complete).pdf

01566-cal-1997 drawings.pdf

01566-cal-1997 form-1.pdf

01566-cal-1997 form-2.pdf

01566-cal-1997 form-3.pdf

01566-cal-1997 form-5.pdf

01566-cal-1997 g.p.a.pdf

01566-cal-1997 priority document.pdf

1566-CAL-1997-(03-10-2012)-FORM-27.pdf

1566-CAL-1997-FORM-27.pdf

1566-CAL-1997-OTHER PATENT DOCUMENT.pdf