METHOD FOR DETERMINING AT LEAST ONE PRODUCTION PARAMETER

Abstract

An inlet screen, arranged at the water inlet of a hydropower plant and comprises a plurality of elongated bars separated by a distance holding means, each elongated bar having in its elongation a proximal portion and a distal portion, and an upstream region and a downstream region, said regions being at an angle in relation to said proximal and distal portions, at least one of said bars defining a space extending along at least a portion of the elongation of said bar, said bar being provided. with an electric heating means. Said elongated bar has an elongated intermediate portion, said space being defined in either of the upstream region and the downstream region, said intermediate portion extending along the elongation of the bar between the upstream region and the downstream region, said electric heating means comprising at least one electric heating member.

Claims

1. A method for determining at least one production parameter of a workpiece to be produced that has amorphous properties, said method comprising the following steps: simulating a cooling rate of a reference workpiece during the production for at least one predetermined alloy and at least one predetermined production method, taking into consideration at least one production method; storing reference simulation data that specify the simulated cooling rate; receiving a component description that specifies a workpiece to be produced, wherein the component description is specified by a data structure, and wherein the component description specifies a geometry of the workpiece to be produced and an alloy; determining a reference component for at least part of the component description, wherein the determination is performed using pattern recognition , wherein the determination of the reference component comprises segmenting the component description into a plurality of component segments, wherein the pattern recognition is performed for at least one component segment; reading out first simulation data for the reference component that specify at least one property of the reference component, wherein the stored reference simulation data are read out as first simulation data; determining at least one production parameter for producing the workpiece using the first simulation data.

2. The method according to claim 1, wherein the segmentation comprises dividing a component geometry specified by the component description, using basic shapes and/or connection points.

3. The method according to claim 1, wherein the determination of the reference component comprises loading a reference data set, in particular from a database unit, wherein the reference data set specifies a plurality of reference components , and wherein the determination comprises selecting the reference component from the reference data set using the pattern recognition

4. The method according to claim 1, wherein the pattern recognition comprises classification, in particular using an artificial neural network, wherein the classification associates a reference component, in particular of a/the reference data set, with component segments , in particular pixels, voxels, volume elements and/or partial segments of the component geometry.

5. The method according to claim 1, wherein the first simulation data specify a cooling rate of the reference component for a predetermined alloy and a predetermined production method.

6. The method according to claim 1, wherein the first simulation data specify mechanical properties of the reference component for a predetermined alloy, a predetermined production method and/or a cooling rate.

7. The method according to claim 1, wherein the simulation of the cooling rate of the reference workpiece during production is performed for for a plurality of predetermined alloys and for a plurality of predetermined production methods, preferably simulation of the cooling rate of the reference workpiece after an injection procedure in an injection molding process; and wherein the storing of the reference simulation data comprises storing in a database unit.

8. The method according to claim 1, wherein determining at least one mechanical property of the workpiece to be produced using a simulated cooling rate of the reference component, wherein the determination of the at least one production parameter is also performed using the determined at least one mechanical property.

9. The method according to claim 1, wherein the simulation of the cooling rate is performed taking into consideration parameters of a production method, for example a stamping speed, an initial temperature and/or a mold geometry.

10. The method according to claim 1, wherein the simulation of the cooling rate or of the mechanical load case is only performed when the stored first simulation data for a reference component do not specify a cooling rate or result data for an identical load case.

11. A production method for producing a workpiece, comprising the following steps: determining at least one production parameter via a method according to claim 1, producing the workpiece using the at least one production parameter.

12. A computer-readable storage medium which contains instructions which prompt the at least one processor to implement a method according to claim 1 when the instructions are executed by the at least one processor.

13. A design system, comprising the following: a receiving unit which is designed to receive a component description that specifies a workpiece to be produced, wherein the component description is specified by a data structure, and wherein the component description specifies a geometry of the workpiece to be produced and an alloy; a determination unit which is designed to determine a reference component for at least part of the component description, wherein the determination unit is designed to perform the determination using pattern recognition, wherein the determination unit is further designed to perform segmentation of the component description into a plurality of component segments, wherein the determination unit is designed to perform the pattern recognition for at least one component segment, wherein the determination unit has a simulation unit which is designed to simulate a cooling rate of a reference component during production for at least one predetermined alloy and for at least one predetermined production method, taking into consideration at least one production method; a database unit that is designed to store first simulation data that specify at least one property of the reference component, and wherein the database unit is further designed to store reference simulation data that specify the simulated cooling rate; a reading unit that is designed to read out the first simulation data for the reference component from the database unit, and wherein the reading unit is further designed to read out the stored reference simulation data; a parameter determination unit which is designed to determine at least one production parameter for producing the workpiece using the first simulation data.

14. The design system according to claim 13, wherein a production machine that is designed to produce the workpiece to be produced using the at least one production parameter.

Description

[0108] The invention is explained in more detail below with reference to exemplary embodiments. In the drawings:

[0109] FIG. 1 shows a schematic illustration of an injection molding machine;

[0110] FIG. 2 shows a schematic illustration of a tool;

[0111] FIG. 3 shows a schematic illustration of a design system;

[0112] FIG. 4 shows a schematic illustration of a determination unit;

[0113] FIG. 5 shows a schematic illustration of a workpiece in a load case;

[0114] FIG. 6 shows a schematic illustration of the segmentation of a workpiece;

[0115] FIG. 7 shows a schematic illustration of the design of an artificial neural network;

[0116] FIG. 8 shows a schematic illustration of a design system in a further exemplary embodiment;

[0117] FIG. 9 shows a flow chart for a method for determining at least one production parameter;

[0118] FIG. 10 shows a flow chart of method steps for segmenting a workpiece to be produced;

[0119] FIG. 11 shows a flow chart for the simulation of a load case;

[0120] FIG. 12 shows a flow chart of a production method.

[0121] In the following, the same reference numbers are used for identical parts or parts having the same effect.

[0122] FIG. 1 shows a schematic representation of an AMM (amorphous metal) injection molding system 1. The injection molding system 1 comprises a mold in the tool 2 and a melting chamber 3. The melting chamber 3 is supplied with a solid alloy segment of an amorphously solidifying alloy (blank) 4 by a robot and is placed centrally in an induction coil 5. The blank 4 is heated within the melting chamber 3 by means of a heating element, in particular an induction field which is generated by the induction coil 5. The blank 4 is a solid alloy segment of an amorphously solidifying alloy. The alloy segment 4 comprises, for example, a certain amount of palladium, platinum, zirconium, titanium, copper, aluminum, magnesium, niobium, silicon and/or yttrium.

[0123] The blank 4 is melted by the heating element or the induction coil 5, so that it is present in molten form. Preferably, the blank 4 is heated to a temperature of 1050° C. The molten material is injected into the tool 2 by a piston 6.

[0124] FIG. 2 shows the schematic design of an injection molding tool 2. The molding chamber 11 is filled with a melt by means of one or a plurality of openings 10 leading into a molding chamber 11 of a tool 2. The molding chamber 11 is designed as a negative form of the workpiece 24 to be produced. In the exemplary embodiment of FIG. 2, it is provided that an opening 10 can be used to guide liquid material into the molding chamber 11. It can be advantageous to use a plurality of sprues for filling the molding chamber 11 in order to achieve a uniform temperature distribution and to reduce turbulence of the melt. A uniform temperature distribution and a small number of turbulences lead to a better cooling operation, to homogeneous cooling and thus to uniform amorphous material properties.

[0125] The liquid material must rapidly cool down within the molding chamber 11 in order to prevent crystallization. The cooling of the liquid material depends greatly on the geometry of the component or workpiece 24 to be produced.

[0126] FIG. 3 shows a schematic representation of a design system 20. The design system 20 comprises a database unit 21, a receiving unit 30, a reading unit 31, a determination unit 32, and a parameter determination unit 33.

[0127] The receiving unit 30 is designed to receive a component description 26 of a workpiece 24 to be produced. In the exemplary embodiment, the component description is implemented as a CAD file. In the shown exemplary embodiment, the workpiece 24 to be produced is designed as a wrench only by way of example. For simplification, it is assumed that the component description 26 specifies a parameterization of the wrench. The component description 26 thus comprises a height parameter H3 which specifies the height or the length of the workpiece 24 to be produced. Moreover, the component description 26 specifies a width B3 that specifies the width of the workpiece 24 to be produced. Of course, further parameters are conceivable but, however, are omitted for simplification.

[0128] In the shown exemplary embodiment, the receiving unit 30 is designed to receive the component description 26 via a communication network. For this purpose, the receiving unit 30 provides a programming interface, for example an API (application programming interface), by means of which user programs may transmit data to the receiving unit 30.

[0129] The receiving unit 30 is communicatively connected to the determination unit 32 so that the component description 26 of the determination unit 32 may be provided by the receiving unit 30.

[0130] The database unit 21 is designed to store component descriptions 25, 25′ of a plurality of reference components 22, 22′. The reference components 22, 22′ may each be associated with different component types. It may thus be seen in FIG. 3 that the reference component 22 is a wrench, and the reference component 22′ is a ratchet combination wrench. The associated component descriptions 25, 25′ likewise respectively specify geometric parameters B1, B2, H1, H2 of the associated reference components 22, 22′.

[0131] Moreover, the database unit 21 is designed to store simulation data 27 associated with the reference components 22, 22′. The simulation data 27 comprise information regarding the production and/or the use of the reference components 22, 22′. In one exemplary embodiment, the simulation data 27 thus comprise cooling rate data for individual segments of the reference components, given production with a selected production method and a selected alloy. The simulation data 27 may additionally or alternatively specify properties of the reference components 22, 22′ derived from the cooling rates. For example, the simulation data 27 specify which component segments of the reference components 22, 22′ have amorphous properties.

[0132] The determination unit 32 is designed to determine a reference component using the component description 26 of the workpiece 24 to be produced. For this purpose, the determination unit 32 may be designed to segment the component description 26. The segmentation deconstructs the component description 26 into component segments. These component segments may, for example, be designed as basic geometric shapes such as cuboids, pyramids or balls.

[0133] The determination unit 32 is additionally or alternatively designed to deconstruct the component description 26 into component segments that each specify individual components. A complex workpiece may thus be deconstructed into sub-workpieces; for example, a screwed-together, complex workpiece may be broken down into the individual workpieces and the screws.

[0134] The determination unit 32 is also designed to determine at least one reference component 22, 22′ using pattern recognition, either for the entire workpiece 24 to be produced or for component segments of the workpiece 24 to be produced. In one embodiment, a type of the workpiece 24 to be produced may first be determined for this purpose. For example, a classifier may be used in order to associate a component type with the workpiece to be produced. Based on the determined component type, the reading unit 31 may read out, from the database unit 21, the reference components 22, 22′ which are associated with the same component type. A preselection may thus be made by means of the component type.

[0135] Alternatively, it is also possible that the reading unit 31 is designed to read out reference components 22, 22′ in succession from the database unit 21 in order to thus check whether a reference component 22, 22′ has sufficient similarity to the workpiece 24 to be produced. A regression system or a classifier may be used to establish the similarity. For example, a classifier may have a binary output that is “1” if sufficient similarity is present and “0” if no sufficient similarity is present. A regression system may specify a value of the similarity, wherein a value of “1” may specify that a full identity is present and a value of “0” may specify that there is no similarity between a reference component 22, 22′ and the workpiece 24 to be produced. The parameters stored in the component descriptions 26, 25, 25′ may also be used for determining the similarity.

[0136] In the shown exemplary embodiment of FIG. 3, the determination unit 32 determines that the workpiece 24 specified by the component description 26 has a high similarity to the reference component 22. The reading unit 31 is designed to read out simulation data 27 of the reference component 22, 22′ that exhibits the greatest similarity to the component 24 to be produced.

[0137] As stated, the simulation data 27 specify, for example, the cooling temperature of the reference component 22 upon production. It may thus be established whether the reference component 22 has amorphous properties.

[0138] The parameter determination unit 33 is now designed to specify, using the simulation data 27, at least one production parameter 28 for the workpiece 24 to be produced. The parameter determination unit 33 is thereby designed to use, for example, certain data of the simulation data 27 as production parameters 28 for the workpiece 24 to be produced. For example, if the simulation data 27 specify that the associated reference component 22 has amorphous properties, the parameter determination unit 33 may thus specify the alloy simulated in the reference component 22 and the simulated production method as production parameters 28 for the workpiece to be produced. It may thus be ensured, without a new simulation, that the workpiece 24 to be produced will also have amorphous properties.

[0139] FIG. 4 shows a schematic view of a determination unit 32. The determination unit 32, in the shown exemplary embodiment, has a classification unit 34 which is designed to determine whether a reference component 22, 22′ is suitable as a basis for determining the at least one production parameter 28 for the workpiece 24 to be produced. As was already stated above with respect to FIG. 3, the classification unit 34 is designed to determine a similarity of the workpiece 24 to be produced and the reference components 22, 22′ stored in the database unit 21.

[0140] Moreover, the determination unit 32 has a simulation unit 35 which is designed to simulate the behavior of the workpiece 24 to be produced or component segments of the workpiece 24 to be produced. The simulation unit 35 is, for example, designed to simulate the temperature behavior during production for a production method and an alloy. It may thus be established whether amorphous properties are achieved with the selected alloy and the production method for the workpiece to be produced, i.e., whether the cooling rate is greater than a critical cooling rate.

[0141] Moreover, the simulation unit 35 is designed to simulate load cases of the workpiece 24 to be produced.

[0142] FIG. 5 thus shows a schematic representation of a load case 50. In the load case 50, a workpiece 24 is attached to a wall 51 at a distal end 52. The workpiece 24 is free at a proximal end 53. The application of a force F perpendicularly from below at the proximal end 53 of the workpiece 24 is now specified by the load case 50. The application of the force F leads to a deformation of the workpiece 24. A torque, i.e., a bending moment, thereby acts on the workpiece 24 so that deformation occurs. If the force F is too great, either irreversible bending or even breakage of the workpiece 24 occurs.

[0143] The simulation unit 35 is now designed to simulate the behavior of the workpiece 24 upon application of the force F. Associated simulation data 27 thus specify how much the workpiece 24 bends and/or whether breakage occurs. The simulation data 27 may thereby also specify a time profile, i.e., the state of the workpiece 24 at different points in time after the application of the force F at the proximal end 53.

[0144] FIGS. 6 to 8 show a further exemplary embodiment of the design system 20.

[0145] FIG. 6 illustrates the segmentation of a workpiece 24 into component segments 29, 29′, 29″. As described, the determination unit 32 is designed to deconstruct a workpiece 24 to be produced into individual component segments 29, 29′, 29″. For this purpose, the determination unit 32 may be implemented as an artificial neural network 40; cf. FIG. 7. In the exemplary embodiment of FIG. 6, the workpiece 24 to be produced is a wrench. The workpiece 24 to be produced is in this case segmented into three component segments 29, 29′, 29″. A first component segment 29 is formed by a first end which comprises first jaws arranged parallel to one another. A second component segment 29′ is formed by the second end of the wrench 24, which has two jaws arranged parallel to one another and is arranged so as to be horizontally mirrored to the first component segment 29. A third component segment 29″ is arranged between the first and the second component segment 29, 29′ and comprises a grip area of the wrench 24.

[0146] In order to segment the workpiece 24 to be produced, the determination unit 32 may process the component description 26 which specifies the workpiece 24 to be produced. For segmentation, a neural network may be used, for example, which associates the individual component segments with a class of objects. For example, the artificial neural network of FIG. 6 may be used for this purpose.

[0147] FIG. 7 shows an artificial neural network which, in the instance of FIG. 6, is designed as a convolutional neural network 40 (CNN 40). The neural network 40 of FIG. 7 can be designed as a classifier or as a regressor.

[0148] Input data 41 for the neural network 40 may be a tensor, i.e., a three-dimensional matrix, which has a plurality of data elements. Each data element can correspond to a volume element. Each data element can be designed as a tuple which indicates whether material is present at the location of the corresponding volume element, what material is used, and/or what initial temperature prevails at the location of the corresponding volume element. The component description 26 of the workpiece 24 to be produced may, for example, be designed as a three-dimensional tensor 41 with volume elements. The component description 26 may thus specify the input data 41 of the neural network 40.

[0149] However, it is also conceivable that the component description 26 is designed as a CAD model which parameterizes basic shapes such as curves, cuboids, spheres and others. The elements of the CAD model may thus also form the input data 41. In this instance, too, when the input data 41 are thus specified by the elements of a CAD model, the elements of the CAD model may be expressed as a tensor in order to form the input data 41.

[0150] A CCN 40 is defined by a plurality of parameters. A kernel sequentially scans the input data 41 in a first step. The stride of the kernel indicates by how many volume elements the kernel must be shifted during each scan. The size of the kernel can also be defined. The stride and the size of the kernel thus define the so-called feature detectors 43 which are generated by a first convolution 42. Each feature detector 43 detects a specific feature in the input data 41. For example, a feature detector 43 may indicate whether or not material is present at a particular location. Overall, a plurality of feature detectors 43 are generated which have not been previously defined manually.

[0151] According to the same principle, a new set of feature generators 45 is generated from the first feature detectors 43 in a second convolution 44, wherein during the second convolution the number of feature generators is reduced compared to the first convolution. Such a step is referred to as pooling or subsampling.

[0152] A third set of feature generators 47 is generated in a third convolution 46. In the last step, a class is assigned to each volume element by means of a so-called soft-max layer. This means that it is apparent from the output as to which class of objects a volume element belongs. As a result, the segmentation of a workpiece 24 to be produced and the association of the workpiece 24 or of individual component segments 29, 29′, 29″ with reference components 22, 22′, 22″ may be performed in a single step, for example by means of the neural network of FIG. 6.

[0153] Each layer of the CCN 40 consists of a large number of neurons, i.e., of activation functions to which weights are assigned. The output of the neuron is activated or not activated depending on the weight and an input value. Possible activation functions include, for example, logit, arc tan, Gaussian functions. Training of the CCN 40 is performed using the backpropagation algorithm, wherein the values of the weights are determined.

[0154] There are a number of different models for CNN, such as VGG-net, RES-net, general adversiral networks or google LeNet. Any of these implementations can be used, or another implementation is possible. Training of the neural networks can be carried out efficiently, since a plurality of the operations can be carried out parallelized. The inference, i.e., the querying of values for certain component description, can be carried out very efficiently.

[0155] FIG. 8 shows an exemplary embodiment of a design system 20 in which a workpiece 24 to be produced is segmented into a plurality of component segments 29, 29′, 29″, and these may be associated with individual reference components 22, 22′, 22″.

[0156] In contrast to the design system 20 of FIG. 1, in the design system 20 of FIG. 8, a component 24 to be produced or the associated component description 26 is received by a receiving unit 30 and segmented by means of the determination unit 32 into individual component segments 29, 29′, 29″.

[0157] The database unit 21 stores at least three reference components 22, 22′, 22″ which correspond to the component segments 29, 29′, 29″. The determination unit 32 is designed to identify the reference components 22, 22′, 22″ corresponding to the component segments 29, 29′, 29″.

[0158] The reading unit 31 subsequently reads out simulation data 25, 25′, 25″ for the identified reference components 22, 22′, 22″. The parameter determination unit 33 thereby determines at least one production parameter 28, based on the read-out simulation data 25, 25′, 25″, and outputs it to an injection molding machine 1. Alternatively, a machine for producing the workpiece 24 to be produced may also be selected with the production parameter 28 in an upstream step.

[0159] In the shown exemplary embodiment, the parameter determination unit 33 is designed to combine the simulation data 25, 25′, 25″ with one another in order to determine the at least one production parameter 28. In this case, the parameter determination unit 33 may use, for example, existing data to simulate a load case for the workpiece to be produced, in order to determine whether the workpiece 24 to be produced withstands the load case on the basis of the simulated behavior of the workpiece 24 to be produced given a predetermined alloy and/or a predetermined production method.

[0160] In one exemplary embodiment, the cooling behavior of the reference components 22, 22′, 22″, which is specified by the simulation data 25, 25′, 25″, may be used in order to establish whether the workpiece 24 to be produced has amorphous properties given an alloy that is used and a production method. In this instance, the simulation data of the simulation data 25, 25′, 25″ may be used as input data of a simulation of the cooling process of the workpiece 24 to be produced.

[0161] The parameter determination device 33 is designed to decide, based on the simulation data 25, 25′, 25″ and/or a simulation of the workpiece 24 to be produced, whether the result of a production with the simulated production parameters corresponds to a requirement profile. If the result of the production does not correspond to the requirement profile, the simulation of the workpiece 24 to be produced is performed with changing parameters until the requirement profile is fulfilled. Only then are the simulated parameters output as at least one production parameter 28.

[0162] In this context, the parameter determination unit 33 may also be designed to perform an optimization of the production parameters with respect to a target value using a cost function. The cooling behavior may thus be used as a cost function which is to be minimized, i.e., the cooling rate is to be maximized. The exceeding of a critical cooling rate which specifies the achievement of amorphous properties may thereby be used as a termination criterion.

[0163] FIG. 9 shows a flow chart for a method 60 for determining at least one production parameter 28, which method is implemented by the design system 20. In a receiving step 61, a component description 26 is received which specifies a workpiece 24 to be produced. In a determination step 62, a reference component 22, 22′, 22″ for the workpiece to be produced is determined, as was already explained in conjunction with FIGS. 3 and 8. For the reference component 22, 22′, 22″, a cooling rate may subsequently be simulated 70. In the simulation 70, reference simulation data 72 are generated which are stored in a database unit 21 in a subsequent storage step 71.

[0164] The reference simulation data 72 and simulation data 27 stored regarding the specific reference component 22, 22′, 22″ are subsequently read out from the database unit 21 in a readout step 63. Using the simulation data 27 or the reference simulation data 72 which may be portions of the simulation data 27, at least one production parameter 28 for the component to be produced is now determined in a determination step 64 according to the principles already described above. In the shown embodiment, the determination 64 of the at least one production parameter 28 comprises selecting 82 a production machine 1 with which the workpiece 24 to be produced is to be produced.

[0165] FIG. 10 shows a flow chart of the use of a determination unit 32. In a first training step 65, a classification and/or regression unit 34 is trained or taught. For this purpose, in the shown exemplary embodiment, the supervised learning method is used. The trained classification or regression unit 34 is used in a segmentation step 66 to segment the workpiece 24 to be produced. The segmentation 66 thereby comprises a dividing step 69 in which the workpiece 24 to be produced is divided into component segments 29, 29′, 29″ (cf. FIG. 6).

[0166] In a pattern recognition step 67, a check is subsequently made, by means of the same or a further classification or regression unit 34, in a pattern recognition step 67, as to which reference component 21, 21′, 21″ stored in a database unit 21 has a sufficient similarity to the workpiece 24 to be produced. A reference component 22, 22′, 22″ is thereby associated with each component segment 29, 29′, 29″ in a classification step 68.

[0167] FIG. 11 illustrates how, in a determination step 80, at least one mechanical property of the workpiece 24 to be produced is determined. The determination 80 is thereby performed using a simulated cooling rate 83. For example, an amorphicity value which may specify the degree of amorphicity may be determined using the simulated cooling rate 83.

[0168] Based on the at least one mechanical property, a load case 50 is now simulated in the subsequent simulation step 81, as is explained in more detail in conjunction with FIG. 5. A check may thereby be made as to whether the workpiece 24 to be produced would withstand the load case 50 with the simulated parameters.

[0169] FIG. 12 shows a flow chart of a production method 90. In a first step 91, at least one production parameter 28 is determined for a workpiece 24 to be produced, as described in detail above. Using the determined at least one production parameter 28, the workpiece 24 to be produced may subsequently be produced by means of a production machine 1.

[0170] It is explicitly pointed out that all described aspects may be combined with one another in any manner. In particular, the aspects described with respect to devices are likewise disclosed for the corresponding methods, and vice versa.

LIST OF REFERENCE SIGNS

[0171] 1 Injection molding machine

[0172] 2 Tool

[0173] 3 Melt cylinder

[0174] 4 Alloying element/blank

[0175] 5 Heating element

[0176] 6 Punch

[0177] 10 Inlet opening

[0178] 11 Molding chamber

[0179] 20 Design system

[0180] 21 Database unit

[0181] 22, 22′, 22″ Reference component

[0182] 24 Workpiece

[0183] 25, 25′, 25″ Component description of a reference workpiece

[0184] 26 Component description of the workpiece to be produced

[0185] 27 Simulation data

[0186] 29, 29′, 29″ Component segments

[0187] 28 Production parameter

[0188] 30 Receiving unit

[0189] 31 Reading unit

[0190] 32 Determination unit

[0191] 33 Parameter determination unit

[0192] 34 Classification unit

[0193] 35 Simulation unit

[0194] 40 Artificial neural network

[0195] 41 Input data/tensor

[0196] 42 First convolution

[0197] 43 Feature detector

[0198] 44 Subsampling

[0199] 45 Second feature detectors

[0200] 46 Second convolution

[0201] 47 Third feature detectors

[0202] 48 Feed-forward layer

[0203] 49 Output layer

[0204] 50 Load case

[0205] 51 Wall

[0206] 52 Distal end

[0207] 53 Proximal end

[0208] 60 Method

[0209] 61 Receiving step

[0210] 62 Determination step

[0211] 63 Readout step

[0212] 64 Determination step of a production parameter

[0213] 65 Training

[0214] 66 Segmentation

[0215] 67 Pattern recognition

[0216] 68 Classification

[0217] 69 Division

[0218] 70 Simulation of cooling rate

[0219] 71 Storage

[0220] 72 Reference simulation data

[0221] 80 Determination step

[0222] 81 Simulating a mechanical load case

[0223] 82 Selection

[0224] 83 Simulated cooling rate

[0225] 90 Production method

[0226] 91 Determining at least one production parameter

[0227] 92 Production

[0228] 93 Selecting a production machine

[0229] F Force

[0230] B1, B2, B3 Width

[0231] H1, H2, H3 Height