EP1364161B1 - Method for the production of a burner unit - Google Patents

Abstract

In the case of swirl-stabilized premix burners ( 1 ), an axial mass flow distribution of the fuel introduced which has especially favorable values with respect to characteristics such as NO<SUB>x </SUB>emission and maximum amplitudes of pulsations occurring is used. For this purpose, Pareto solutions are determined with respect to the said characteristics, in that a distributing device ( 5 ) with control valves is represented by a tree structure with distributing parameters, and values for the distributing parameters on the basis of which the distributing device ( 5 ) is set by means of a control unit ( 10 ) are iteratively generated in a data-processing system ( 9 ) by an evolutionary algorithm. On the basis of the values determined by a measuring unit ( 11 ), solutions which are especially favorable with respect to the characteristics mentioned, espectially Pareto-optimal, are selected. The distributing devices or the premix burners of the burner system are then formed in a way corresponding to such a solution.

Description

Technical area

The invention relates to a method for producing a burner system according to the preamble of claim 1. Such burner systems are mainly used in gas turbines. Such a procedure is over DE 4 446 945 A1 known.

State of the art

It is known that generic burner systems with conventional spin-stabilized premix burners, in which the fuel is usually introduced more or less evenly distributed over the length, in various respects have problematic properties that have to do with the end of the combustion. Above all, the exhaust gases often contain a considerable proportion of pollutants, in particular NO _x . Also caused by pulsating combustion pressure waves often cause difficulties because they expose the gas turbine high mechanical loads and reduce their life.

To alleviate these problems, it has been proposed to stabilize the combustion by means of feedback by influencing the pressure in the burner system. For this purpose, the pressure was measured there and fed the measured signal out of phase via speakers again. In this way, a more stable combustion and thus a reduction in the formation of pressure waves and also the NO _x and CO emissions could be achieved. See CO Paschereit, E. Gutmark, W. Weisenstein: 'Structure and Control of Thermoacoustic Instabilities in a Gas Turbine Combustor, Combust. Sci. and tech. 138 (1998), pp. 213-232 , However, the required equipment is very considerable.

Presentation of the invention

The invention is based on the object to provide a method for producing generic burner systems, which are simple and in which the combustion proceeds favorably, in particular with regard to the reduction of pulsations and low emissions of pollutants, in particular NO _x . It has been found that the combustion process is strongly influenced by the mass flow distribution of the fuel introduced into the premix burners.

According to the invention, the burner systems are designed so that the fuel is introduced with a certain mass flow distribution in the premix burner, which ensures favorable combustion properties, in particular with regard to pulsations and pollutant emissions.

Brief description of the drawings

Fig. 1

schematisch einen Vormischbrenner mit einer vorgeschalteten Verteilvorrichtung,

Fig. 2

schematisch einen Aufbau einer Versuchsanlage mit einem Vormischbrenner entsprechend Fig. 1 und einer Verteilvorrichtung sowie einer Datenverarbeitungsanlage zur Ermittlung günstiger Massenstromverteilungen,

Fig. 3

ein Schema einer Baumstruktur als vereinfachtes Modell für die Massenstromverteilung;

Fig. 4, 5a,b

allgemein das bei der Ermittlung günstiger Massenstromverteilungen eingesetzte Optimierungsverfahren, wobei

Fig. 4

die Bestimmungsmenge eines typischen Optimierungsproblems und ihre Abbildung auf die entsprechende Zielmenge zeigt und

Fig. 5a,b

Schritte bei der Auswahl neuer Bestimmungsgrössen aus vorher erzeugten Versuchsgrössen im Zielraum;

Fig. 6a,b

den Zielraum des vorliegenden Optimierungsproblems nach 20 bzw. 64 Iterationsschritten,

Fig. 7

Massenstromverteilungen gemäss ausgewählten Lösungen des Optimierungsproblems.

In the following the invention will be explained in more detail with reference to figures, which represent only one embodiment. Show it

Fig. 1

schematically a premix burner with an upstream distributor,

Fig. 2

schematically a construction of a pilot plant with a premix burner accordingly Fig. 1 and a distribution device and a data processing system for determining favorable mass flow distributions,

Fig. 3

a schematic of a tree structure as a simplified model for mass flow distribution;

Fig. 4, 5a, b

in general, the optimization method used in the determination of favorable mass flow distributions, wherein

Fig. 4

the determination set of a typical optimization problem and its mapping to the corresponding target set shows and

Fig. 5a, b

Steps in the selection of new determinants from previously generated experimental quantities in the target area;

Fig. 6a, b

the target area of the present optimization problem after 20 or 64 iterations,

Fig. 7

Mass flow distributions according to selected solutions of the optimization problem.

Ways to carry out the invention

A premix burner 1 ( Fig. 1 ) basically known structure, as used in a burner chamber of a gas turbine, has the shape of a truncated cone with at its wide end an outflow opening 2. Along two diametrically opposite generatrices air inlet slots 3a, b are provided, on whose outer sides in each case 16 inlet openings 4 for the fuel supply are arranged, which form the burner-side end points of a distribution device 5.

In the course of the production of a burner system, mass flow distributions are determined for the time being, which are as favorable as possible with regard to a target variable whose components are formed by certain properties, in particular the emission of NO _x and the maximum of the amplitudes occurring pressure surges. This is done by means of a test setup ( Fig. 2 ), in which one as in connection with the Fig. 1 described trained premix burner 1 is preceded by a suitable for experimental purposes distribution device 5, for example, as in Fig. 1 can be shown formed.

The input of the distribution device 5 is formed by a supply line 6, which is connected to a fuel source, for. B. a stationary gas line (not shown) connected and provided with an input valve 7 which limits the fuel supply. Subsequently, the main line 6 branches into two branch lines 8a, b, from each branch off four supply lines, in each of which a control valve is located. The control valves are designated V ₁ to V ₈ . Subsequent to the respective control valve, the supply line branches to two pairs of opposing inlet openings 4 in such a way that two axially successive groups of four of inlet openings are each acted upon via one of the control valves V ₁ , ..., V ₈ with fuel. The control valves V ₁ ,..., V ₈ are designed so that certain mass flows m ₁ ,..., M ₈ can be set with them. The two arranged on the same side inlet openings 4 each upstream of an on / off valve. By means of the on / off valves V " ₁ , ..., V" ₁₆ can in each case the fuel supply to two successive inlet openings 4 are selectively blocked.

The structure of the distribution device 5 may differ from what has been described in many respects. Thus, each control valve may be associated with a larger or smaller group of inlet openings or even a single inlet opening. The on / off valves can be used elsewhere or even omitted or it can only be used such valves, for. B. one for each inlet. The topology may be different, for. B. the in Fig. 3 illustrated distribution device 5 '( Fig. 3 ), a tree structure of three-way valves, as described in more detail below. The experiments, the results of which are given below, were carried out with a distributor device similar to that described in US Pat Fig. 1 shown, but without the on / off valves V " ₁ , ..., V" ₁₆ .

The control valves V ₁ ,..., V _{8 of} the distribution device 5 are set by a control unit 10 according to values output by the data processing system 9. A measuring unit 11 supplies the measured characteristics of the burner installation to the data processing installation 9. For the representation of the mass flow distribution in the data processing installation 9, the distribution apparatus 5 is placed on the distribution apparatus 5 '(FIG. Fig. 3 ), ie using a model in which it is represented by a binary tree structure of three-way valves V ' ₁ , ..., V' ₇ and assuming that the total mass flow has a fixed value M in each case. The position of each of the three-way valves can be represented by a distribution parameter p, 0≤p≤1, which is the fraction falling to the left output in the distribution of the mass flow between the left and the right output equivalent. If the individual mass flows at the outputs of the control valves V ₁ ,..., V _{8 are designated} m ₁ ,..., M ₈ , the distribution parameter of the valve V ' ₁ is p ₁ = (m ₁ + .. . + m ₄ ) / M, that of the valve V ' ₂ to p ₂ = (m ₁ + m ₂ ) / (m ₁ + ... + m ₄ ) etc. and vice versa, one can use the distribution parameters p ₁ ,. .., p ₇ easily m ₁ , ..., m ₈ compute according to m ₁ = Mp ₁ p ₂ p ₄ , m ₂ = Mp ₁ p ₂ (1-p ₄ ) etc .. By virtue of the fact that the data processing system 9 with In the case of the model described above, only seven parameters are required, which reduces the dimension of the determination space (see below) by one.

If, as in the present case, it is optimized for several independent properties, it is generally not possible to select a particular optimal solution, but a lot of so-called Pareto optimal solutions can be found, each characterized by being not Pareto-dominated , d. H. that there is no other solution which would be more favorable in terms of a property and unfavorable in respect of any of the other properties. In other words, a solution that is more favorable in terms of at least one property than a Pareto-optimal solution is inevitably less favorable than at least one other property.

The Pareto optimal solution objectives usually form a hypersurface section in the target space spanned by the target variables, the so-called Pareto front, which is the target set, ie the set of target values of all possible solutions against areas of the target area which would be more favorable but inaccessible, bounded. The Pareto front is followed by other hypersurfaces that border the target, which solutions Although not Pareto-optimal, but u. U. are still of interest.

For the search for Pareto-optimal solutions offer semistochastic methods, which z. For example, they can be oriented towards the natural evolutionary process of living beings through crossing, mutation and selection and realized by means of so-called evolutionary algorithms. With the help of the same, starting from certain, z. For example, randomized output quantities for a set of determinants iteratively approximates Pareto-optimal solutions by determining the determinants z. B. varied by recombinations and random mutations and is selected from the trial sizes thus prepared by based on the corresponding target variables selection a new set of determinants. As soon as a certain abort criterion is met, the iteration is aborted.

In Fig. 4 For example, a situation is illustrated in which the determination space is 3-dimensional with parameters x ₁ , x _2, and x ₃ . The determination set B, over which the determination variable varies, is limited by the fact that the variables each lie between zero and an upper limit X ₁ , X ₂ or X ₃ and therefore forms a cuboid, the product of the intervals [0, X ₁ ], [0, X ₂ ] and [0, X ₃ ]. By a known functional relationship f, which can be given by a mathematical model or by a test setup, each determination variable x = (x ₁ , x ₂ , x ₃ ) is assigned a target variable y = f ( x ), which in a target quantity Z lies. It is a subset of the 2-dimensional target space in this case, ie, y = (y ₁ , y ₂ ), where y ₁ and y _{2 represent} two properties to be optimized. The target quantity Z can be the complete image quantity of the Determination set B under the figure f or a constrained part of the same.

The target values of the solutions sought form a so-called Pareto front P (solid line), which bounds the target set Z against small, ie favorable values of the properties y ₁ , y ₂ . Laterally, P solutions are connected to the Pareto front, which also lie on the edge of the target set Z. They are not Pareto-optimal, since for each of these solutions a solution can be found on the Pareto front, where both properties are cheaper, but they can u. U. also be of interest.

It is now primarily a question of finding determinants x in which the associated target values y = f ( x ) are as close as possible to the Pareto front P. In addition, they should be fairly uniformly distributed over the entire Pareto front P and if possible also over the border areas of the target set Z adjoining the same. Such solutions are generated by means of an iterative evolutionary or genetic algorithm, which forms the basis of a program running on a data processing system. In this case, each variable is usually a bit vector of a length L, the z. B. 32 is coded.

In order to find approximately Pareto-optimal solutions, first output quantities lying in the determination set B are generated, which form the starting point of the iteration as the first set of determinants. You can z. B. be distributed regularly or randomly over the determination set B. Then so many iteration steps are executed until a termination criterion is met. It may be that a certain maximum of Iterationsschritten executed or a certain computation time was consumed or even in that the change of the target sizes during a certain number of iteration steps has remained below a certain minimum.

• Rekombination: Es werden jeweils durch Kombination von Teilen mehrerer Bestimmungsgrössen aus dem vorliegenden Satz neue Grössen erzeugt. Beispielsweise werden zuerst entweder sämtliche möglichen geordneten Paare von Bestimmungsgrössen gebildet oder auch nur einige mittels eines Zufallsgenerators bestimmte. Jede Bestimmungsgrösse bildet einen Vektor aus n reellen Parametern. Es wird nun ebenfalls mittels eines Zufallsgenerators eine Zahl 1 mit 0≤l≤n erzeugt und dann zwei neue Grössen gebildet, indem die ersten l Parameter von der ersten Bestimmungsgrösse und die übrigen von der zweiten Bestimmungsgrösse genommen werden und umgekehrt.
• Mutation: Zu den im Rekombinationsschritt erzeugten Grössen werden mittels eines Zufallsgenerators z. B. gemäss einer Normalverteilung erzeugte Grössen addiert. Natürlich können aus einer Grösse auch mehrere Ausgangsgrössen auf solche Weise erzeugt werden.
• Selektion: Aus den beiden obengenannten Schritten geht eine Menge von Versuchsgrössen hervor, die in der Regel grösser als der ursprüngliche Satz von Bestimmungsgrössen ist. Aus dieser meist verhältnismässig grossen Menge von Versuchsgrössen wird nun ein neuer Satz von im Durchschnitt besonders günstigen Bestimmungsgrössen ausgewählt. Die Vorgehensweise bei der Selektion ist für die Entwicklung der Iteration von grosser Bedeutung. Zur Steuerung der Annäherung an die Pareto-Front P und an angrenzende Bereiche des Randes der Zielmenge Z, insbesondere zur Erzielung einer breiten Annäherung, wird vorzugsweise wie folgt vorgegangen:

At each iteration step, the following substeps are performed:

• Recombination: New quantities are created by combining parts of several determinants from the given set. For example, either all possible ordered pairs of determinants are first formed, or only a few are determined by means of a random number generator. Each determinant forms a vector of n real parameters. A number 1 with 0≤l≤n is now likewise generated by means of a random generator and then two new quantities are formed by taking the first parameters from the first determination variable and the remaining ones from the second determination variable, and vice versa.
• Mutation: To the sizes generated in the recombination step by means of a random generator z. B. summed according to a normal distribution generated quantities. Of course, from a single size, several output quantities can be generated in such a way.
• Selection: The two above-mentioned steps produce a set of experimental quantities, which is usually larger than the original set of determinants. From this usually comparatively large amount of experimental quantities, a new set of on average particularly favorable parameters is selected. The The selection procedure is very important for the development of the iteration. To control the approach to the Pareto front P and to adjacent areas of the edge of the target set Z, in particular to achieve a wide approach, the following procedure is preferably used:

In a first selection step, the hyperplane designated by the condition y ₁ = 0 of a part of the target space which comprises the target quantity Z and that in the 2-dimensional case shown in FIG. Fig. 5a ) coincides with the y ₂ axis, subject to a partition in subsets, which in this case form intervals I ₁ ⁱ . On the basis of this, the said part of the target space is subdivided into subsets W ₁ ⁱ , which are the archetypes of the orthogonal projections thereof along the positive y ₁ axis to the said intervals I ₁ ⁱ . In other words, the subset W ₁ ⁱ for a given i is the set of all points y = (y ₁ , y ₂ ) in said part of the target space for which y ₁ > 0 and y _{2 is} in I ₁ ⁱ . In Fig. 5a it forms a parallel to the coordinate axis y ₁ strip.

For each of the non-overlapping subsets W ₁ ⁱ , that test variable is now determined and selected for which y _{1 is} optimal, ie minimal. In Fig. 5a If the target quantities of all test variables are marked with a circle O, the test variables selected in the individual W ₁ ⁱ are marked with a superimposed mark x.

In a second selection step, the part of the target space containing the target set Z is subdivided into subsets W ₂ ^j in a very analogous manner and there again, the test variable is determined and selected for each subset for which y _{2 is} optimal, ie minimal. These solutions are in Fig. 5b marked with an overlaid plus sign +. The new set of determinants with which the next iteration step is then tackled consists of the test variables selected in both selection steps.

In comparatively many cases, especially near the middle area of the Pareto front P, it is the same experimental quantities that are determined in both cases, so that a selection step is usually sufficient to establish these experimental quantities. However, in the lateral edge regions and especially in the part of the edge of the target quantity Z which adjoins the Pareto front P, this is generally not the case. If one also values the determination of solutions in these ranges, then it is necessary to carry out both selection steps.

Of course, it is also possible to select in each of the subsets each not only a trial size, but a selected set of trial sizes, z. For example, k is the most favorable with respect to the remaining component with k> 1.

The described procedure in the selection can easily be transferred to cases in which the dimension m of the target area is greater than 2. In this case, one will preferably form all m hyperplanes, which are characterized in that one of the coordinates y ₁ ,..., Y _{m is} equal to zero and each perform a partition thereof in subsets. This can be done in such a way that from the outset each of the coordinate axes is subdivided into intervals, and as subsets of a hyperplane then all products of intervals are used in each case which are subdivided the coordinate planes spanning the hyperplane.

In each of the subsets which are formed by the archetypes of the orthogonal projections on the subsets of the hyperplanes, then the most favorable test variable with regard to the remaining component is selected, and finally the combination of the selected test variables over the subsets and hyperplanes is formed to produce the new set of parameters. Depending on whether one is interested in finding the most comprehensive possible solutions or, above all, in determining certain areas, the selection can only take into account a part of the hyperplanes, especially, as explained above, the central areas of the pareto. Front are usually already well captured at the first selection step.

The concrete procedure determined by the algorithm can, of course, deviate from the above description by a different summary of individual steps, etc., in particular, it is not absolutely necessary for the described selection steps to be carried out one after the other.

The subdivision into intervals can be graded either uniformly or logarithmically, but it can also be finer in areas where there is a particular interest. The partitions in subsets may be held or altered throughout the iteration, e.g. B. be adapted to the distribution of the target sizes. Instead of or next to hyperplanes, it is also possible to use subspaces of a lower dimension, but then it has to be optimized in each subset with regard to several properties, which requires further specifications or a recursive procedure.

Thus, in the selection various modifications of the described approach are conceivable. The procedure described has the advantage that the determination of the solutions is controlled in each case by the specifications with respect to the position of the target variables such that the target variables derived from them are finally distributed in a desired manner over an edge region of the target quantity. Of course, various modifications are possible in recombination and mutation. These sub-steps are not necessarily required in both cases.

In the present optimization problem, the determination space is spanned by the distribution parameters p ₁ ,..., P ₇ , which in each case can vary over the interval [0,1], the target space by emissions and pulsations, in the example the two properties NO _x content and maximum amplitude A of the occurring pressure waves. The target area is in Fig. 6a, 6b shown with the target values of the 100 solutions determined after 20 iteration steps ( Fig. 6a ) and the 320 solutions determined after 64 iterations ( Fig. 6b The two figures clearly show how, above all, favorable solutions are being identified and the limit of the set of target values against the favorable values of the properties - the Pareto front - is beginning to emerge.

From the determined solutions, a specific solution is selected, whereby further, possibly more intuitive criteria can be included in the decision. The determination variable of the selected solution is then based on the production of the burner system, in particular the production or adjustment of the distributor 5. So it is made a burner system in which the Distribution parameters p ₁ , ..., p ₇ and thus the mass flows m ₁ , ..., m _{8 were set} so that they correspond to the size of the selected solution.

Contains the distribution device 5 in addition to the control valves also, as in Fig. 1 shown on / off valves, the determination space must be supplemented by corresponding binary switching parameters, each represented by a bit that can take the values 0 for 'closed' and 1 for 'open'. The occurrence of these parameters changes practically nothing in the optimization process described above. Only with the mutation a change is necessary. Here can z. B. be provided that each switching parameters, so each bit, with a certain, z. B. fixed probability is inverted, so 0 goes into 1 and 1 in 0.

Fig. 7 shows as examples five different solutions, ie mass flow distributions, the abscissa showing the numbers of the control valves V ₁ , ..., V ₈ and the ordinate the mass flows m ₁ , ..., m ₈ . The properties thus obtained are shown in the following table: solution character NO _x content [ppm] Maximum amplitude [mbar] 1 (equal distribution) circles 2.5 3.12 2 diamond 3.0 2.92 3 triangle 4.0 2.83 4 cross 5.0 2.80 5 square 2.0 3.37

Solutions 3 and 4 offer particularly favorable values with regard to the pressure surges that occur, while solution 5 is the best exhaust gas values shows, however, with high values for the pressure maxima. Solution 2, on the other hand, offers again very good properties in this respect, for which only a slightly increased NO _x emission must be accepted.

Of course, various deviations from the example described are possible. Thus, additional properties or other than those described such. B. CO emissions, average amplitude of the generated sound u. a. to be based on the optimization. Also, the optimization method may differ from what has been described. It is also possible to carry out the search for Pareto-optimal solutions at different loads and corresponding values of the total mass flow M and thus to find solutions which have favorable properties over a larger working range.

Finally, the solution which best suits the requirements is selected and a burner plant is produced in which the respective premix burners used in the test arrangement each have a fixed axial mass flow distribution which corresponds to the size of the solution chosen. The adjustment of the desired mass flow distribution can be done in various ways. So z. B. in the burner system distribution devices with throttles or switches are used, which produce the desired solid mass flow distribution in the simplest and most reliable way. However, the mass flow distribution can also be very easily adjusted by the dimensions, in particular the Diameter of the inlet openings. In this case, the Distributor each consist of a pipe system that connects their input with the inlet openings.

LIST OF REFERENCE NUMBERS

1

premix

2

Opening

3a, b

Air inlets

4

inlet openings

5

distributor

6

main

7

inlet valve

8th

branch lines

9

Data processing system

10

control unit

11

measuring unit

Claims (10)

1. Method for producing a burner system with a fuel source and at least one swirl-stabilized premix burner (1), and also a distributing device (5), by means of which a number of inlet openings (4) in the premix burner (1) are connected to the fuel source, characterized in that a desired mass flow distribution into the at least one premix burner (1) is determined, in that a number of determination variables fixing the mass flow distribution, vectors from a determination set which is a subset of an n-dimensional domain, are generated one after the other by means of a data-processing system (9) and, in a test setup with at least one premix burner (1) and at least one adjustable distributing device (5), the mass flow distribution is respectively set on the basis of the determination variable, and a target variable, a vector from a target set which is a subset of an m-dimensional domain, is measured on the test setup and, finally, a suitable determination variable is selected on the basis of the target variables and the at least one premix burner or the at least one distributing device of the burner system is designed such that the mass flow distribution corresponds to that which is fixed by the selected determination variable.
2. Method according to Claim 1, characterized in that the components of the determination variables are formed at least partly by the distributing parameters (p₁, ..., p₇) of the branching points of a tree structure, by which the distribution of the mass flow between inlet openings or groups of inlet openings (4) of the at least one premix burner (1) is determined.
3. Method according to Claim 1 or 2, characterized in that Pareto solutions, which are distinguished by the fact that in every solution in which one component of the target variable has a more favorable value, at least one other component inevitably has a less favorable value, are determined at least approximately by means of the data-processing system (9) and the suitable determination variable among the Pareto solutions is selected.
4. Method according to Claim 3, characterized in that the approximative determination of the Pareto solutions is performed in that starting variables serving as a set of determination variables are determined and, until a terminating criterion is satisfied, the data-processing system (9) carries out iteration steps in which a new set of determination variables is in each case determined from a set of determination variables by generating from the set of determination variables a set of test variables respectively lying in the determination set, from which the new set of determination variables is selected in each case on the basis of the target variables which were measured for the mass flow distribution fixed by the determination variables.
5. Method according to Claim 4, characterized in that the generation of the test variables from the set of determination variables is performed by random mutation or recombination of the determination variables by means of the data-processing system (9).
6. Method according to one of Claims 1 to 5, characterized in that the concentration of at least one pollutant, for example the NO_x concentration, in the exhaust gas forms a component of the target variable.
7. Method according to one of Claims 1 to 6, characterized in that a measure of the pulsations occurring in the burner system, preferably their maximum amplitude, forms a component of the target variable.
8. Method according to one of Claims 1 to 7, characterized in that the inlet openings (4) are provided at least partly axially in succession.
9. Method according to one of Claims 1 to 8, characterized in that the desired mass flow distribution is achieved at least partly by dimensioning of the inlet openings (4).
10. Method according to one of Claims 1 to 9, characterized in that the desired mass flow distribution is achieved at least partly by restrictors and/or diverters of the distributing device.

Images