EP0951977A1 - Method and apparatus for controlling and optimizing technical processes, especially for the manufacture of fibreboards - Google Patents

Abstract

The process control and system, to optimize the mfg. process in the production of fiberboards, has a model for at least one quality function together with the associated cost function. A total quality function is produced to show the user which quality meets the min. requirements. At least one model is formed for at least one quality function including bending strength, water absorption, chemical oxygen demand, and module of elasticity. They give the total quality function together with the associated cost functions. The bending strength and water absorption are taken for special evaluation in the total quality function. A total quality function is formed for an optimum process. The total quality function is a total of the separate values, especially a total of models weighted with sigmoid-quadratic functions and computed production costs. The models for quality forecasting are produced by data-driven modeling methods or otherwise by computer processing. Recommendations are given on the best costings for given quality functions. The total quality function is optimized by computer processing. The control values which are produced are checked, and validated, for the computed optimizing of the total quality function. An Independent claim is included for an apparatus with a monitor screen (60) to operate the production assembly, and especially the refiner to convert the wood material into fibers. Preferred Features: The monitor screen (60) has at least one input field (61) and a field (63) for the results with an interactive working after insertion of the nominal data and the expected results. A functional display (62), especially of a refiner, is between the input (61) and results (63) fields.

Description

The invention relates to a process control method and process optimization of technical processes, a suitable one to optimize the manufacturing quality Working point is determined with the individual quality functions various quality parameters of the product to be manufactured considered. In particular, the invention relates to the application of the process for the production of fiberboard. In addition, the invention also relates to a associated device using a computer.

In industrial practice, process control and Process optimization in the large-scale production of Products play an increasing role. By using Process control can be used as far as possible in control systems automate and perform at suitable work points. In addition to automation, the optimization of Quality of manufacture of the products to be manufactured an essential Criteria.

Holz" und/oder Altpapier" über verschiedene Prozeßschritte beispielsweise Papier als Endprodukt hergestellt wird. Ein wichtiger Teilschritt ist dabei die Herstellung von Holzfasern im sog. Refiner, einem Mahlwerk, zwischen dessen gegenläufig rotierenden Scheiben das Holz in seine einzelnen Fasern zerrieben wird. Aus einem derartigen Holzfaserstoff können auch Faserplatten hergestellt werden, die wegen der ökologischen Eigenschaften zunehmende Anwendung in der industriellen Praxis finden. The latter applies, for example, to the pulp and paper industry, where raw materials are used

Wood "and / or Waste paper "is produced in various process steps, for example paper as the end product. An important sub-step is the production of wood fibers in the so-called refiner, a grinder between whose counter-rotating discs the wood is ground into its individual fibers. Fiberboard can also be made from such wood fiber material are produced, which are used increasingly in industrial practice due to their ecological properties.

It is already known from the general prior art to set up quality functions for various quality parameters for products to be manufactured. The magazine publishes a special application for chipboard and fiberboard production Wood as raw and material "55 (3) (1997), pages 133 to 140 shows that a suitable process modeling can lead to online quality control and process optimization. It is usually assumed that the production costs increase with the quality of the products to be manufactured.

Based on this, the object of the invention is a method of the type mentioned in such a way that at a production quality specified by the user be minimized.

The object is achieved in that a Model for at least one quality function and equally that associated cost function can be determined and that an overall quality function is determined, the user stipulating what quality must be achieved as a minimum requirement. At the invention is therefore always an optimizing overall quality function formed, for which the models for quality prediction based on a data analysis.

• Biegefestigkeit,
• Wasseraufnahme,
• Chemischer Sauerstoffbedarf (CSB),
• Elastizitätsmodul,
  und daß daraus zusammen mit den zugehörigen Kosten die Gesamtgütefunktion ermittelt wird. Vorzugsweise werden die Güte-funktion für die Biegefestigkeit und die Wasseraufnahme, welche für die Herstellung von Faserplatten besonders charakteristische Größen sind, bestimmt und in die Gesamtgütefunktion geeignet bewertet eingebracht.

The method according to the invention is preferably used in the pulp and paper industry. It is particularly suitable for process control and process optimization in the production of fiberboard, whereby models are determined for at least one of the quality functions relevant for the fiberboard:

• Bending strength,
• Water absorption,
• Chemical oxygen demand (COD),
• Modulus of elasticity,
  and that the overall quality function is determined from this together with the associated costs. The quality function for the flexural strength and the water absorption, which are particularly characteristic sizes for the production of fiberboard, are preferably determined and appropriately evaluated in the overall quality function.

With a suitable device for performing the invention Procedure with a computer, being on a The associated data can be displayed on the monitor Monitor equally the user interface for operating a Plant, in particular a refiner for the production of the Fibrous material from wood.

Figur 1

ein Optimierungsmodell in der Anwendung zur Herstellung von Faserstoff aus Holz,

Figur 2

das bei Figur 1 zugrunde gelegte Refiner-Modell,

Figur 3

eine Darstellung einer vorteilhaften Bewertung von zwei Gütekriterien,

Figur 4

eindimensionale Darstellungen zur Verdeutlichung der Auswahl der Optimierungsfunktion für die Gesamtgüte,

Figur 5

eine Darstellung mit der Datenbasis, dem daraus entstandenen Modell und den erhaltenen Kurven, die eine Aussage über die Zuverlässigkeit der Optimierungsergebnisse erlauben,

Figur 6

eine Bedienoberfläche zur Realisierung des Verfahrens der Berücksichtigung von Gütefunktion einerseits und Kostenfunktion andererseits.

Further details of advantages and advantages of the invention result from the following description of the figures of exemplary embodiments with reference to the drawing in conjunction with the patent claims. Show it

Figure 1

an optimization model in the application for the production of fibrous material from wood,

Figure 2

the refiner model used in FIG. 1,

Figure 3

a representation of an advantageous evaluation of two quality criteria,

Figure 4

one-dimensional representations to clarify the selection of the optimization function for the overall quality,

Figure 5

a representation with the database, the resulting model and the curves obtained, which allow a statement about the reliability of the optimization results,

Figure 6

a user interface for realizing the method of taking into account quality function on the one hand and cost function on the other hand.

In FIG. 1, 10 denotes a refiner model with inputs and outputs, where detailed below, 20 one unit for quality assessment, 30 one unit for Cost assessment and 40 a summation unit for linking of the functions from units 20 and 30 an overall function to be minimized.

For example, the refiner model has six inputs 1 to 6 and four outputs 7 to 9 and 11 for each characteristic Sizes.

The refiner model 10 enters via inputs 1 to 6, for example a signal 1 from the I-refiner, a signal 2 from the I-Pregrinding-Refiner, a signal 3 from the number of revolutions the transport screw, a signal 4 from the vapor pressure Signal 5 from the wood age and a signal 6 from the used Wood mix.

For example, the refiner model 10 uses as outputs Signal 7 for the flexural strength, a signal 8 for the Water intake, a signal 9 for energy costs and a Receive signal 11 for raw material costs. Signals 6 and 7 arrive at unit 20 of the quality assessment and the Signals 9 and 11 for the unit of the cost assessment.

In Figure 2 it is shown that the signals 1 to 6 in different Calculation units 12 to 14 as part of the Refiner model arrive. The unit 12 is used for cost calculation a so-called energy function, the unit 13 for Preprocessing and calculation of company sizes for the Refiner and the unit 14 for cost calculation of the raw materials. For example, using suitable fuzzy models 15 can be a prediction of quality from the refiner parameters enable. The results are signals 7, 8, 9 and 11 of Figure 1.

It can be sufficient to optimize sales, only two suitable ones To use model quality functions, each by a Model resulting from a data analysis are determined. For example, one model good describes the one from the Production with given parameters resulting bending strength and the other resulting water absorbency.

The user sets the desired maximum water absorption capacity and the minimum flexural strength. About that He also determines how strong these requirements are Quality must be taken into account if it is in its entirety cannot be met. The overall quality function to be optimized consists of the evaluations of the model quality function and the hardboard produced per ton incurred costs for the raw materials, the electrical energy and the required steam together.

The overall quality function is determined to be the two Model quality functions and production costs are included. It a multicriteria quality function is thus formed. In order to can be the goal of optimization, the cost of production of hardboard taking into account the boundary conditions to minimize the quality of the panels achieved become. Accordingly, the overall quality function corresponds to Sum of the calculated production costs and an evaluation the quality properties calculated by the model quality function the hardboard.

The evaluation assigns production parameters, by means of which the quality requirements are not met, to an energy which increases quadratically with the increase in the degree of violation of the requirements. Production parameters, by means of which the requirements are met, are assessed by a uniform low energy, so that the minimum of the overall quality function in this area depends exclusively on the calculated production costs. If quality requirements are set that cannot be met by any production parameters, the optimum lies in the quadratic range of the evaluation. The position of the optimum can be varied by parameter α ₁ (see below) so that the individual requirements are taken into account to different degrees. The user can therefore choose between the individual requirements that cannot be met in their entirety weigh ".

FIG. 3 clearly shows how for two criteria a combined evaluation function can look like. Here are in the coordinates of the as X1 and X2 with two criteria Measured numbers from 0 to 100 are given, which differ for themselves Have quality functions and a common quality assessment 31 linked, which is the result of unity 20 reproduces from Figure 2.

It has been shown that specifically for fiberboard manufacture a sigmoid-square evaluation practical Delivers results. This is shown for two criteria of Figure 3 by the quality assessment function 31 as a level with sufficient goodness and adjoining it rising edges with deteriorating quality.

The general form of the sigmoid-quadratic function for an n-dimensional input space is given as equation (1) in the appendix with the parameters h ₁ , h _o , α ₁ , α _o , d _i , p, σ _i , k.

• d_i [d₁ = -1, d₂ = 1]: Ausrichtung der Funktion: Für d₁ = 1 bzw. = -1 ist die Funktion monoton steigend bzw. fallend in χ_i und gibt damit eine obere bzw. untere Schranke für den Parameter χ_i an; iE {1,2, ..., n}.
• σ_i[σ₁ = 1,σ₂ = 1]: Skalierung des Wertebereichs der i-ten Dimension zur Normierung der Modellausgaben; iE {1,2, ..., n}.
• h_o, h₁[h_o = 100, h₁ = 1000]: Der Parameter h_o gibt das Niveau der nahezu konstanten Ebene an, während h₁ das Niveau angibt, ab dem der quadratische Verlauf einsetzt. Die Höhe des Absatzes" ist daher h₁ - h_o.
• k[k = 1]: Gibt die Steilheit des Absatzes" an. Kleinere Werte bewirken einen steileren Verlauf.
• p[p = 2]: Mit diesem Parameter kann die initiale Steigung der quadratischen Funktion am Übergang beeinflußt werden. Unter der Annahme eines unendlich steilen Absatzes (k → 0) ist die Ableitung der Bewertungsfunktion in Richtung χ_i am Übergang der entsprechenden Variablen gerade
  
  iE{1,2, ..., n}. Zu kleine Werte für p können die Konvergenz wegen des resultierenden flachen Verlaufs der Energiefunktion verlangsamen.
• α_o, α_i, α₂ [α_o = 50, α₁ = 50, α₂ = 50]: Über die Faktoren α_i können die quadratischen Funktionen skaliert werden, wobei die Skalierungsfaktoren durch αo = 1 n i=1 n α i normiert werden. Diese Faktoren sind nur von Bedeutung, wenn die durch die Grenzen χ_i,max/min gestellten Forderungen insgesamt nicht erfüllbar sind, also zum Beispiel die geforderte maximale Wasseraufnahmefähigkeit in Kombination mit der geforderten minimalen Biegefestigkeit nicht erreicht werden kann. Die zu den durch Variation der Prozeßparameter erzielbaren Merkmalskombinationen gehörende Punktmenge in der Eingabemenge der sigquad-Funktion liegt in diesem Fall vollständig außerhalb des durch die χ_i,max/min begrenzten Gebietes. Durch die Skalierung der quadratischen Funktionen kann nun die Lage des lokalen Minimums der Bewertungsfunktion auf dieser Punktmenge beeinflußt werden. Dabei wird im Optimierungsergebnis die Forderung an den Parameter χ1 um so stärker berücksichtigt, je größer α1 gegenüber den übrigen Gewichtungen ist; iE {1,2, ..., n}.
• s[s = 2]: Dieser Exponent bestimmt die Sensitivität der Verschiebung des lokalen Optimums bei Variation der Skalierungsfaktoren α₁; iE {1,2, ..., n}.

The course of the function can be influenced via the parameters, the values given in brackets being used in FIG. 3:

• d _i [d ₁ = -1, d ₂ = 1]: Alignment of the function: For d ₁ = 1 or = -1 the function is monotonically increasing or decreasing in χ _i and thus gives an upper or lower limit for the parameter χ _i ; iE {1,2, ..., n}.
• σ _i [σ ₁ = 1, σ ₂ = 1]: scaling the range of values of the i-th dimension to normalize the model outputs; iE {1,2, ..., n}.
• h _o , h ₁ [h _o = 100, h ₁ = 1000]: The parameter h _o indicates the level of the almost constant plane, while h ₁ indicates the level from which the quadratic course begins. The amount of Paragraph "is therefore h ₁ - h _o .
• k [k = 1]: Indicates the slope of the Paragraph ". Smaller values result in a steeper course.
• p [p = 2]: This parameter can be used to influence the initial slope of the quadratic function at the transition. Assuming an infinitely steep paragraph (k → 0), the derivation of the evaluation function in the direction χ _{i is} straight at the transition of the corresponding variable
  
  iE {1,2, ..., n}. Values for p that are too small can slow down the convergence because of the resulting flat course of the energy function.
• α _o , α _i , α ₂ [α _o = 50, α ₁ = 50, α ₂ = 50]: The quadratic functions can be scaled using the factors α _i , the scaling factors being represented by α O = 1 n i = 1 n α i be standardized. These factors are only important if the requirements set by the limits χ _{i, max / min} cannot be met overall, for example the required maximum water absorption capacity in combination with the required minimum bending strength cannot be achieved. In this case, the number of points in the input quantity of the sigquad function belonging to the combinations of features which can be achieved by varying the process parameters lies completely outside the area delimited by the χ _{i, max / min} . The position of the local minimum of the evaluation function on this point set can now be influenced by scaling the quadratic functions. In the optimization result, the requirement for parameter χ1 is taken into account the greater the larger α1 is compared to the other weightings; iE {1,2, ..., n}.
• s [s = 2]: This exponent determines the sensitivity of the shift of the local optimum with variation of the scaling factors α ₁ ; iE {1,2, ..., n}.

FIG. 4 demonstrates the influence of the variation of some of these parameters on a one-dimensional sigmoid-quadratic function with the parameter values d = 1, σ = 5, h _o = 0, h ₁ = 100, k = 0.5, p = 3, α _o = 2, α ₁ = 2 and s = 10.

In Figure 4, the influence of a Parameters reproduced in one dimension: is shown the change in the one-dimensional function 41 upon variation one parameter at a time, which clearly shows that influences many properties of the function isolated from each other can be and thus the function to various Requirements can be adjusted. Upper and lower limits curve 41 are provided with reference numerals 42 and 43.

To calculate the optimum of the overall quality function different, known mathematical methods, such as B. Gradient methods or stochastic methods, be used. In particular, genetic Algorithms are used.

In order to guarantee a certain quality of the optimization result regardless of the convergence of the optimization methods and the chosen starting point, the overall quality is at a finite number of points of a grid in the input space a priori "and the point with the best quality is used as the starting point. Since the calculation effort for this is very high, it is possible to generate a definable number of random samples in the input room, of which those with the lowest resulting value of the total quality function as the starting point for an optimization process is used. This results in simple scalability and, in the case of multiple executions, an already calculated optimization result can be successively improved. This simple method for starting point selection is only made possible by using the sigmoid-square evaluation. If the boundary conditions are not integrated in the energy function, With very strict or even unachievable boundary conditions, the problem of determining a starting point of a complexity similar to that of solving the optimization problem itself can be this, in particular if the boundary conditions are not stated analytically can be, so that contradictions in the boundary conditions can not be determined. When using the sigmoid-quadratic energy function, any starting point in the search area can in principle be selected for optimization, since the energy function automatically leads the optimization, as far as possible, into the permissible areas.

The validation strategy is illustrated in FIG. Here it is essential whether there are many in the area of the solution found Entry points of the database, on the basis of which the model quality function arose. For this, the Value of a corresponding density function at the location of the Optimization result calculated. There are many data points in the area of the optimal point found, the local modeling errors as a deviation between the model and the real data in the environment of the optimization result can be determined in the form of a standard deviation. This Validation strategy thus enables the assessment of the practical relevance of the optimization results. Such Assessment is especially with data-driven. Models like they were used for refiner optimization, important because the amount and quality of the modeling Data generally available in various Can strongly differentiate areas of the parameter space. The Assessment of the density and spread of the underlying Data also provides immediate information for others Tests to be carried out on the system to be optimized.

Based on FIG. 5, a statement about the quality can thus be made of the calculated optimum, where in one abstract example using graph 51 the model output, who clarified when specifying one using the crosses Database was obtained is shown. Lines 52 and 53 illustrate the measurement of the local deviation.

The optimization method described with reference to FIGS. 1 to 5 is largely software-based on a computer Monitor 60, which also serves as a user interface, carried out. In Figure 6 it is clarified how an implementation for example to operate a refiner. On the user interface 60 there is an input field 61, a Functional representation 62 of the refiner and a parameter and Result field 63 reproduced. After entering the by the user required quality data for the fiberboard as a product determine suitable production parameters for the refiner and displayed in field 63.

In input field 61, the input parameters of the optimization with the desired quality index are identified by field 610: For example, the desired minimum flexural strength can be displayed in field 611, which can be, for example, 30,000 N / mm ² . The required maximum water absorption (field 612) water uptake "), which can be specified, for example, with 70%. Corresponding values for the weighting of bending strength and water absorption can be specified on the displays 612 and 613. The user can enter minimum and maximum sizes by mouse input or the like The optimization can then be started by clicking on label 615.

On the one hand, the output parameters appear on the display 63 for the refiner according to field 630, on the other hand the resulting quality and cost parameters accordingly Field 640 and finally a confirmation or validation corresponding to field 650. In detail, the Optimization according to field 631 a certain operating current for the main grinder of the refiner and accordingly Field 632 a certain flow for the pre-grinder of the Refiners and continue its predetermined rotation speed corresponding to field 633. Additional specifications for the Wood age in years (a's) and proportion of hard and softwood corresponding to fields 634 to 636 and further specifications for steam pressure according to field 637.

Using the confirmation display 650 with the fields 651 to 655, validation data and status can be displayed. Fields 651 and 652 show the local density of the database for the model of the bending strength or the heat absorption capacity in the environment of the optimization result according to field 630. In fields 653 and 654 the local Scattering of the database is shown in the optimization result.

Field 640 of display 63 is out of use the parameters resulting from field 630 resulting quality values of the produced wood pulp specified: For example, the Fields 641 and 642 show the actual values of flexural strength and water uptake and fields 643, 644 and 645 the Costs of the necessary amounts of electrical energy, of steam energy and wood. You can do this in the confirmation screens the local deviations in bending strength and Water absorption can also be displayed.

Through an interactive way of working, users can Specifications G1 vary and are adapted to current needs become. It is essential that the result in terms of quality and costs can be read directly on the user interface.

The process described was exemplified specifically for use in a refiner in the pulp and paper industry, in particular for the production of fiberboard. The optimization process shown can, however, also be used for other technical processes in which products are produced and the costs are to be minimized taking into account the quality.

Claims (12)

Process for process control and process optimization of technical processes for the manufacture of products, a suitable working point being determined to optimize the quality of manufacture, which takes into account various quality parameters of the product to be manufactured and the associated production costs with individual quality functions, characterized in that one model each for at least one quality function and likewise the associated cost function is determined, and that an overall quality function is determined, the user specifying which quality must be achieved as a minimum requirement. A method according to claim 1, characterized in the use in the production of fiberboard from wood pulp produced by a refiner, a model being determined for at least one of the following quality functions of the fiberboard to be produced: Bending strength, Water absorption, Chemical oxygen demand (COD), Modulus of elasticity,
and that the overall quality function is determined from this together with the associated costs. Method according to claim 2, characterized in that the quality functions for the flexural strength and the water absorption are determined and appropriately incorporated into the overall quality function. Method according to Claim 1, 2 or 3, characterized in that an optimizing overall quality function is formed. Method according to Claim 4, characterized in that the overall quality function is formed as the sum of individual quantities, in particular as the sum of model qualities weighted by sigmoid-quadratic functions and calculated production costs. Method according to claim 1 or claim 2, characterized in that the models for quality prediction are created by data-driven modeling methods or are otherwise available with computer support. Method according to one of the preceding claims, characterized in that requirements for the model qualities to be taken into account when optimizing costs are specified. Method according to one of the preceding claims, characterized in that the optimum of the overall quality function is determined by means of suitable computing methods. Method according to claim 8, characterized in that after determining suitable control variables, the calculated optimum of the overall quality function is checked and validated. Device for carrying out the method according to Claim 1 or one of Claims 2 to 9, with a computer and associated monitor, the associated data being able to be displayed on the monitor, characterized in that the monitor equally controls the user interface (60) for operating the system, especially for a refiner for the production of fibrous material from wood. Device according to Claim 10, characterized in that the user interface (60) has at least one input field (61) and one result field (63), the results to be expected being shown by an interactive method of operation after input of solidates. Apparatus according to claim 11, characterized in that a functional representation (62) of the process means, in particular a refiner, is arranged between the input field (61) and the result field (63).

Images