Generation of optimized process variable values and control data for additive construction processes

Abstract

A method and apparatus (60) for generating an optimized process variable value (PGO) for an additive construction process of a raw product (2, 2 ', 2' ') is described. For this purpose, requirement data (AD) of the raw product (2, 2 ', 2' ') is provided. The optimization method is then carried out in order to determine an optimized process variable value (PGO) taking into account the requirement data (AD), at least one optimized scan direction distribution (SSV) for at least one region of the raw product (2, 2 ', 2' ') being determined as the optimized process variable value (PGO) using a KI-based optimization unit (NN, NPS, NSV, NNW, KNSP, KNWS). An optimized process variable value (PGO) is then provided. Furthermore, a method and a control data generation device (54, 54 ') for generating control data (BSD, PSD), a method for creating a KI-based optimization unit (KNSP, KNWS), a control method and a control device (50) for an additively constructed production plant (1) and a corresponding production plant (1) are described.

Description

Technical Field

[0001] The invention relates to an optimized process variable value for generating or determining an additive construction process for a product (Fertigungsprodukts) (hereinafter also referred to as a "component") The invention also relates to a method and a device for generating control data for a production device for the additive construction of at least one product in an additive construction process, as well as a method and a control device for controlling a production device for the additive construction of a product. The invention also relates to a method for creating a KI-based optimization unit, such as a neural network that can be used in one of the above-mentioned methods. The invention also relates to a production device for the additive construction of a product in an additive construction process, having at least one such control device. Background Art

[0002] "Additive construction processes" (also called "additive manufacturing processes") are playing an increasing role in prototype production but also in series production. In general, "additive construction processes" are to be understood as construction processes in which a product is constructed by depositing material ("construction material") based on 3D digital construction data. The construction is therefore usually, but not necessarily, carried out in layers. The term "3D printing" is often used as a synonym for additive construction, the production of models, samples and prototypes by additive construction processes is often referred to as "rapid prototyping", and the production tools are referred to as "rapid tooling".

[0003] The basic method for realizing additive construction processes consists in the selective solidification of the construction material, wherein in many manufacturing processes this solidification can be carried out with the aid of radiation energy, for example electromagnetic radiation, in particular light and / or heat radiation, but if necessary also with the aid of particle radiation, for example electron beam radiation. Such methods using radiation are also referred to as "laser melting". Examples of this are the so-called "laser powder bed melting process" (also called "selective laser sintering" or "selective laser melting") or the "electronic powder bed melting process". In this process, thin layers of the construction material, which are mainly in powder form, are repeatedly superimposed one on top of the other, and in each layer, the construction material is selectively solidified in a "welding process" by irradiating spatially defined locations that become part of the finished product after processing, i.e. the powder particles of the construction material are partially or completely melted by the energy introduced locally at these locations by the radiation. After cooling, these powder particles bond together to form a solid.

[0004] When curing the building material, the energy beam is guided along a predetermined scanning path, usually taking into account a defined irradiation strategy, mainly a so-called "filling strategy", within the contour of the area to be cured in the layers above the layer on the building field, so as to melt and cure the material in the desired spatial and temporal sequence. In addition, further process parameter values ​​(Prozessparameterwerte), such as intensity, focal range or energy beam expansion (Energiestrahlausdehnung) (such as energy beam diameter), intensity distribution shape (or intensity profile) and feed rate (or scanning speed) of the energy beam, layer thickness, etc. are also specified and should be adhered to as strictly as possible.

[0005] Recent research results have shown that during additive construction, certain process variables have a significant influence on the microstructure that is locally formed in the component. Not only, but especially, the metal as the construction material is affected in this way. The microstructure in turn determines the component properties at the macro level and thus the component quality, in particular whether it meets certain quality requirements. As will be described later, the key process variables include in particular the filling strategy in addition to the process parameter values ​​of the energy beam. In addition, all these process variables also have an impact on the construction speed and thus on productivity, energy consumption and construction costs. Therefore, when optimally setting certain process variables, i.e. selecting appropriate process variable values, it may be necessary to balance conflicting objectives (e.g. construction speed on the one hand and stiffness or strength of the component on the other hand).

[0006] Similarly, in other additive construction processes, such as processes in which the material is applied only at the desired location by means of a material application head and subsequently solidified or solidified, various process variables, in particular the choice of the material solidification path (hereinafter such solidification paths are often also referred to as "scanning paths") and the feed rate, can have a considerable influence on the component properties and quality of the component on the one hand and on the productivity on the other, which is why the values ​​of the process variables must be selected wisely. This also applies in principle to additive construction processes such as powder deposition welding (laser cladding) and wire deposition welding (direct energy deposition (DED) or arc additive manufacturing (WAAM)). Summary of the invention

[0007] It is therefore an object of the present invention to provide a suitable method for generating optimized process variable values ​​for an additive construction process and generating control data based on the process variable values ​​or for additive construction of a product, as well as a suitable device for this purpose.

[0008] Therefore, the methods for achieving this purpose include: a method for generating optimized process variable values ​​according to claim 1, a method for generating control data according to claim 11, a method for controlling a production device for additive construction of a product according to claim 12, a method for creating a KI-based optimization unit according to claim 13, a device for generating optimized process variable values ​​according to claim 14, a control data generating device according to claim 15, a control device for a production device for additive construction of a product according to claim 16, and a production device for additive construction of a product according to claim 17.

[0009] The method according to the invention for generating or determining optimized process variable values ​​for an additive construction process (or manufacturing process) of a product consisting of a multilayer construction material has at least the following method steps:

[0010] First, the required data of the product are provided. These required data include, for example, the geometric data of the product. In the simplest case, the geometric data may only be the maximum size and / or minimum size determined by the available construction space, for example. However, the geometric data may also include certain precise dimensions, such as the dimensions of the sub-components or parts of the component, such as the dimensions of the connector, so as to connect the component with other sub-components, the length of the component accurately maintained in certain extension directions, etc. In particular, the geometric data may also include the precise dimensions and all details of the component. The geometric data may be provided in any manner, such as by inputting the geometric data in a user interface, by receiving the geometric data from other program segments, networks and / or data storages. For example, the geometric data may also include component CAD data, which may be taken from a design program, for example. However, the required data may especially include data on other requirements, such as mechanical stress requirements and heat treatment requirements, etc. If no special heat treatment is required after the component is manufactured, the heat treatment requirements are, for example, only the heat requirements obtained by cooling the component, especially the cooling rate. The heat treatment requirements are usually defined by a function having a time-temperature curve (i.e., a temperature-time variation function). In many cases, for example when the component does not require special measures for cooling after manufacturing, these thermal treatment requirements (or the corresponding functions defining these thermal requirements) can also be omitted in the optimization method.

[0011] Then, in the method according to the invention, an optimization method for determining optimized process variable values ​​is performed taking into account the requirement data. Here, at least one optimized scanning direction distribution is determined for at least one region of the product as an optimized process variable value using a KI-based optimization unit. Here, "KI-based" means that the optimization unit is based on artificial intelligence (KI). This can be, for example, a neural network, which will be explained in more detail later with reference to examples.

[0012] Such a "region of a product" can be, for example, a certain (virtual) section of a component, which, as will be explained later, preferably spans multiple layers. As will also be explained, a component can be (virtually) divided into so-called "segments", wherein a segment here preferably includes sub-segments / regions of the product. The sum of the segments of the product then constitutes the product. However, it is also possible, especially in the case of small items, that the entire product consists of only one segment. More complex components usually have multiple segments. In principle, the "region of the product" itself can also include multiple segments or any segments, or the entire product.

[0013] The distribution of scanning directions within a segment is also referred to below as the "segment scanning direction distribution" because it is the distribution of scanning directions within the segment. Thus, the segment scanning direction distribution indicates the frequency of occurrence of various scanning directions (e.g., scanning direction angles) in the relevant segment. "Scanning" is generally understood here to mean the movement of a unit responsible for solidifying the material at each position along a specified "scanning path", such as a material application head that distributes the material and solidifies it, and / or an energy beam used for solidification, etc. For example, in the beam melting process described at the beginning, "scanning" refers to the movement of the energy beam impact point (i.e., the movement of the laser focus in selective laser melting and similar processes) on the current working surface along a specified "scanning path". The current "scanning direction" refers to the current direction along the currently traveling scanning path. For example, the scanning direction can be given by the scanning direction angle of the current working surface with respect to an arbitrarily specified reference direction. The movement speed of the impact surface of the energy beam or the unit responsible for solidifying the material at each position of the construction field is the scanning speed, which can also be adjusted according to the position, i.e., it does not have to be kept constant. "Working surface" generally refers to a plane that is perpendicular to the component construction direction at the corresponding point. In the above-mentioned "laser powder bed fusion process", this refers to the plane in which the powder layer is applied, i.e. the scanning path of the layer usually lies in a plane that is not tilted during the solidification of the layer. For other additive construction processes such as powder deposition welding (laser cladding) and wire deposition welding (direct energy deposition (DED) or arc additive manufacturing (WAAM)), the working surface can also be defined by a so-called tangential plane, but this does not limit its generality. The origin of such a tangential plane is the point of impact of the radiated energy on the material.

[0014] It should be mentioned here that the scanning path does not have to extend continuously, but can also include a plurality of scanning path portions arranged at intervals from each other, in particular in one plane. Therefore, the single "filling line" also described below, along which the energy beam moves on the material layer of the working surface according to the "filling direction arrangement" (generally also referred to as "filling strategy") to solidify the component cross section on the plane, can each be regarded as a separate "scanning path portion".

[0015] As mentioned above, in the laser melting process, the selective irradiation of the energy beam or the displacement of the impact surface in the construction field is usually carried out according to an appropriate irradiation strategy, as mentioned above. Usually, during the solidification process, larger two-dimensional areas, i.e. larger areas of the construction field, are irradiated. Regardless of how the energy beam is generated and regardless of how the impact point is displaced in the construction field, it has proven to be advantageous to initially virtually "divide" at least such larger area to be irradiated according to a selected pattern, for example into virtual "stripes", diamond patterns, checkerboard patterns or the like. Then the individual surfaces of this pattern, i.e. defined sub-areas, such as geometrically standardized surface areas such as stripes or grids, are formed. Usually it is scanned by the energy beam in the form of so-called "shadow lines" (also commonly called "filling lines"). In the case of a stripe pattern, the construction material (from a macroscopic point of view) gradually solidifies along strips parallel to each other, while in detail (from a microscopic point of view) the impact surface of the energy beam moves along shadow lines that are closely attached to each other in the construction field, and the shadow lines extend back and forth within the boundaries of the irradiated strips transversely to the extension direction of each irradiated strip. The filling direction arrangement or filling strategy can be defined, for example, whether it is to work with alternating filling directions (alternating irradiation) or with a constant filling direction (unidirectional irradiation, i.e. jumping back from the end of one filling line in the irradiated strip to the starting point of the next adjacent filling line). Therefore, the filling direction can also be regarded as a set of local scanning directions. In the contour area of ​​the component, the scanning path usually extends along the contour to make the surface as smooth as possible.

[0016] The above-mentioned "scanning direction distribution" (or "segment scanning direction distribution" for a segment), as will be described in detail below, depends, among other things, on the "layer scanning direction arrangement" selected during the construction process. This "layer scanning direction arrangement" generally defines the basic strategy for the operation of the scanning paths during the layered construction, i.e. the irradiation strategy for the individual layers in the case of beam melting, i.e. in what manner or direction the scanning paths are to be operated relative to one another in the layer, and if necessary, in what order the scanning paths are to be operated in the layer, in order to melt and solidify the material in the desired spatial and temporal sequence. Thus, the "layer scanning direction arrangement" defines the relevant scanning directions to be or have been assigned for the main part of the layer surface during the layer-by-layer construction process. As with the filling strategy mentioned above, the layer scanning direction arrangement can also, in general, have a significant influence on the locally formed microstructure in the component as a process variable. It should be noted here that a rotation of the orientation of the layer scanning direction arrangement from layer to layer (described in detail below) is not to be understood as a change in the layer scanning direction arrangement. In other words, even if the orientation changes (rotation around the main construction direction of the layer-by-layer stacking), the layers can be considered to be created with the same layer scanning direction arrangement. Changes in individual scan path sections, in particular changes along the component contour of each layer, such as changes caused by changes in orientation or changes in the component contour from layer to layer, are not considered to be major changes in the layer scan direction arrangement in this sense. That is, in the sense of the present invention, the layer scan direction arrangement of each layer can be considered to be the same, because such changes usually do not lead to major changes in the "in-layer scan direction distribution" (which is basically determined by the layer scan direction arrangement) and therefore do not lead to major changes in the attribute values ​​of the segment. Therefore, typical examples of "layer scan direction arrangement" include the aforementioned fill direction arrangement or fill strategy or they can also be defined by the layer scan direction arrangement.

[0017] Therefore, the scanning direction distribution is advantageously an optimization variable in the optimization method. It is preferably a continuous, particularly preferably fixed optimization variable in the optimization method. Likewise, the scanning direction distribution can also be defined as "quasi-continuous", for example by a sufficient number of discrete values. Thus, a "quasi-continuous" definition of the segment scanning direction distribution can be performed, for example, on a plane formed by a sufficient number of discrete adjacent values, for example 360 ​​support points with an angular range of 360°.

[0018] Preferably, as will be explained in more detail later, in the optimization method, other optimized process variable values ​​can also be determined, such as optimal parameter sets for components, in particular optimal parameter sets for individual segments of a component. Such "parameter sets" for example each include a defined group of process parameter values, i.e. a tuple of individual process parameter values, which are then used to control or optimally control the machine in order to construct at least one layer of the relevant segment.

[0019] The optimized or best process variable values ​​determined in the optimization method (e.g. optimized scanning direction distribution(s), in particular segment scanning direction distribution and, if applicable, the best parameter set) are ultimately provided in order to generate optimized control data based on them, for example, for controlling the production device during the construction process. The provision of the optimized process variable values ​​may include, for example, storage for later use and / or transfer to another computing unit and / or transmission to the production device.

[0020] Depending on the specific design, the method according to the invention for generating or determining optimized process variable values ​​enables a very general optimization of the property profiles (Eigenschaftsprofils) of additively constructed components and is advantageously not limited to the optimization of individual component properties, such as mechanical strength. On the contrary, it represents the possibility of solving boundary value problems for any thermophysical and manufacturing technology properties. In addition to taking into account the requirements profile (based on the requirements data), the most cost-effective way to achieve the specified requirements in terms of production technology can also be determined within the framework of the proposed method according to the design. For example, this goal can be achieved by maximizing the volume construction rate, as will be described later.

[0021] Examples of suitable optimization methods, such as optimization methods using segment scanning direction distributions as optimization variables and selecting an optimal parameter set, have been described in detail in the patent application DE 10 2022 117 935, to which reference may be made, the content of which is hereby fully incorporated. In this regard, the present invention may in particular be based on the method mentioned in this application or further improve the method mentioned therein. In addition to the optimization method in this document, according to the present invention, a KI-based optimization unit is also used here to find at least one optimized scanning direction distribution, such as (multiple) segment scanning direction distributions, for at least one area of ​​the product. This significantly speeds up the entire method, as will be described later with the aid of an exemplary method flow. In particular, in this way, using the KI-based optimization unit, a certain preselection of optimized process variable values, in particular (multiple) scanning direction distributions, can be performed within a single stage of the entire optimization method, thereby simplifying and accelerating the subsequent process.

[0022] Based on the optimized process variable values ​​according to the invention obtained in this way, control data according to the invention can then be generated for a production device for additive construction of at least one raw product.

[0023] A corresponding method according to the invention for generating control data for a production device for additive construction of at least one product consisting of a multilayer construction material comprises at least the following method steps:

[0024] In a first phase, optimized process variable values ​​generated for an additive construction process according to a method in accordance with the above-described features of the invention are provided, for example directly adopted or retrieved from a memory.

[0025] - In a second stage, control data for the production device are then generated in such a way that the optimized process variable values ​​are fully achievable in the additive construction process according to a predetermined evaluation criterion and are preferably also maintained in the manufacturing process. According to current technical possibilities, the optimized process variable values ​​may not be maintained exactly or can only be maintained with great effort. Therefore, the predetermined evaluation criterion should preferably be defined in such a way that the optimized process variable values ​​are achieved as well as possible or approximately (the optimal value is the exact value) in the production process or at least approximately exist within a predetermined tolerance range of the corresponding process variable value and are maintained in the manufacturing process. The tolerance range can also depend on the corresponding optimized process variable value.

[0026] Preferably, the control data of the production device (i.e., the design of the production device should also be adapted thereto), with which, as described at the outset, a building material, preferably a powder, is structured and selectively solidified in a laser melting process, preferably based on a powder bed, wherein, for solidification, the building material is irradiated by at least one energy beam on a building field, wherein the impact surface of the energy beam is moved on the building field along a predetermined scanning path in order to melt the building material in a target zone in and around the impact surface. The "movement" of the energy beam or the impact surface of the energy beam is to be understood here as a general deflection of the energy beam, for example by a galvanometer mirror, but also as a movement of a complete radiation-emitting unit, for example in the form of a diode array, in particular a laser diode array, or by a moving beam shaper. The "target zone" here refers on the one hand to the impact zone, i.e., the zone where the energy beam impacts the surface, and on the other hand also to the area below, i.e., deep into the material or layer, and possibly also to the environment around the impact surface, where the energy beam still acts, for example, by heat conduction in the building material. For the sake of completeness, it should be reiterated that the energy beam can be both particle radiation and electromagnetic radiation, for example light radiation or preferably laser radiation.

[0027] Thus, the control data may preferably be exposure control data, such as scanning data defining or specifying the movement of the energy beam on the surface, control data for setting the energy level or laser intensity, data on the beam "shape" or beam profile and / or beam focus and expansion perpendicular to the beam direction. In addition, as will be described later, the control data may also include other control information, such as coating control data specifying the current coating thickness, information for controlling preheating or postheating using other energy input media, information for injecting inert gas, etc.

[0028] It should also be mentioned here that the control data can be used not only for "simple" process control but also for process regulation, for example by specifying target data for further regulating the process via the control data. In other words, the method according to the invention can also be used to derive the variables required by a regulator, which receives, for example, actual data for feedback, which are determined by melt pool monitoring or time-resolved and / or spatially-resolved imaging, such as thermal imaging by optical tomography, for monitoring the formation layer. Such methods are known to those skilled in the art. In this way, disturbances occurring in the production process can be corrected in order to achieve the target process control specified by the control data as close as possible.

[0029] In the method according to the invention for controlling a device for the production of additive construction of a product, control data are first generated in the manner mentioned according to the invention and then used to control the device via the control data. The control data can be generated in advance and transmitted to the device as a complete data packet or a certain "control protocol", which then executes the production process. In principle, however, the control data for the next layer or segment can also be determined during an already running process, for example when a layer or segment is being solidified.

[0030] According to the invention, in the optimization method, at least one KI-based optimization unit is used. Such a KI-based optimization unit should generally be created in a manner adapted to the optimization method.

[0031] For example, at least one KI-based optimization unit preferably comprises at least one reinforcement learning-based optimization unit and / or a neural network (in particular a deep learning network). If it is a neural network, it must first be trained in an appropriate manner. Only a trained neural network can subsequently form a KI-based optimization unit suitable for the method.

[0032] A method according to the invention for creating a KI-based optimization unit (e.g. for constructing and training a neural network), which method can be used in the above-mentioned optimization method to determine optimized process variable values ​​for a plurality of different types of requirement data (in particular requirements or requirement parameters, such as mechanical stress requirements and thermal treatment requirements), comprises at least the following method steps:

[0033] First, at least one first optimization unit based on KI is determined (e.g. a first neural network is trained), which determines optimized process variable values ​​of a first type of process variables (e.g. an optimal parameter set) based on first type of requirement data (e.g. mechanical stress requirements).

[0034] - Secondly, at least one second optimization unit based on KI is determined (for example, training a second neural network), the second optimization unit determines the optimized process variable value of the second type of process variable (for example, scanning direction distribution) based on the first type of requirement data (for example, the above-mentioned mechanical stress requirement), or the second optimization unit determines the optimized process variable value of the first type of process variable (for example, optimal parameter set) or the second type of process variable (for example, scanning direction distribution) based on the second type of requirement data (for example, heat treatment requirement).

[0035] - Then, a KI-based combined optimization unit (e.g., a combined neural network) is created using a training method in which at least the first KI-based optimization unit and the second KI-based optimization unit are combined with each other to monitor the training of the KI-based combined optimization unit.

[0036] In this way, KI-based (combined) optimization units can be created with lower computational and time costs (compared to training using only classical (i.e. not KI-based) optimization methods, but these optimization methods are also feasible in principle, as will be shown later by way of example) which can quickly determine optimal values ​​for different process variables of different types or different types of combinations of different process variables based on arbitrarily various different types of requirement data.

[0037] It is pointed out again that the above-mentioned requirement data or process variables serve only as examples for the method according to the invention for creating a KI-based optimization unit, although in this specific case these requirement data or process variables are preferably used data or variables within the desired optimization scope.

[0038] The device according to the invention for generating or determining optimized process variable values ​​for an additive construction process for producing a product has at least the following components (for carrying out the method according to the invention explained above):

[0039] - A requirement interface unit, which is designed to provide requirement data for the product, which requirement data include, for example, geometrical data of the product. This can be, for example, an interface for receiving data and / or a memory for storing the data.

[0040] an optimization unit, which is designed to carry out the above-mentioned optimization method for determining optimized process variable values ​​taking into account the requirement data, the optimization unit comprising at least one KI-based optimization unit in order to determine at least one optimized scanning direction distribution as optimized process variable value for at least one area of ​​the product,

[0041] - a process variable value interface unit, which is designed to provide an optimized process variable value, including scanning direction distribution(s), e.g. optimal segment scanning direction distribution(s) and an optimal parameter set. This can be, for example, an interface for transmitting data and / or a memory for storing these data. In principle, the required interface unit and the process variable value interface unit can also be implemented as a common unit, or at least use common components, such as a common memory.

[0042] The control data generating device according to the present invention is used to generate control data of a production device for additive construction of a product in an additive construction process, preferably in the above-mentioned laser melting process, comprising at least the following components:

[0043] - a device for generating or determining optimized process variable values ​​for an additive construction process for producing a product according to the invention as described above and / or an interface of such a device for receiving optimized process variable values. The interface may also access a memory, such as a database, in which the optimized process variable values ​​are pre-stored by the device for generating the optimized process variable values.

[0044] - A data generation unit for generating control data for the production device in such a way that optimized process variable values ​​in the additive construction process are adequately achieved according to predetermined evaluation criteria, as explained above in conjunction with the method for generating control data.

[0045] The control data generating device can be, for example, a part of a control device of this production device for producing the additive structure of the product. However, it can also be implemented independently on another computer in order to subsequently transfer the data to the control device.

[0046] Thus, the control device of the production device for additive manufacturing of a manufacturing process according to the invention has a control data generating device according to the invention and / or an interface of this control data generating device for receiving relevant control data from the control data generating device. The interface in turn includes the possibility of accessing a memory, such as a database, in which the control data are pre-stored by, for example, the control data generating device. The control device is designed to control the production device using this control data, for example to irradiate the building material with an energy beam.

[0047] The production device according to the invention for additively constructing a product in an additive construction process or manufacturing process has, in addition to the usual components depending on the type of manufacturing process, such as for (preferably) laser melting processes, a supply device for introducing construction material, for example in the form of a construction material layer, into the processing area and an irradiation device for selectively solidifying the construction material by irradiation with the aid of an energy beam, and at least one such control device.

[0048] It should be noted here that the device can also have a plurality of irradiation devices, which are then controlled in a coordinated manner via control data in order to fully achieve an optimized process variable value according to a given evaluation criterion or to maintain the process variable value during the manufacturing process.

[0049] The device according to the invention for generating or determining optimized process variable values ​​and the control data generating device according to the invention can both be realized primarily in the form of a computer unit, each individually or in the form of a shared computer unit, and equipped with suitable software. The computer unit can, for example, have one or more cooperating microprocessors or similar devices. In particular, the microprocessor can be realized in the form of suitable software program segments in a computer unit of a control device of a production device according to the invention. The advantage of a primarily software-based implementation is that previously used computer units, in particular control devices of production devices for additive construction, can be easily retrofitted by software or firmware updates so that they work in accordance with the invention.

[0050] In this respect, the object is also achieved by a corresponding computer program product having a computer program that can be directly loaded into a computer unit, in particular directly loaded into a storage device of a device for generating or determining optimized process variable values, a control data generating device or a control device, and having program sections so that all steps of the method according to the invention can be performed when the computer unit or the control device executes the computer program. In principle, the required software components or program sections can also be distributed on a plurality of computer units networked with one another, in which sense the computer units can also be regarded as a common, merely distributed computer unit.

[0051] In addition to the computer program, such a computer program product may also include other components, such as documentation and / or other components, and hardware components, such as a hardware key (dongle, etc.) for using the software. Computer-readable medium, such as a memory stick, hard drive or other removable or permanently installed data carrier, on which program segments of a computer program that can be read and executed by a computer unit, in particular a control device, are stored, which can be used for transmission to a computer unit or a control device and / or for storage on or in a computer unit or a control device.

[0052] In particular, the method according to the invention for creating a KI-based optimization unit can also be implemented with the aid of a computer program product, which can be installed on any computer. The KI-based optimization unit (e.g. a neural network) can also itself form a computer program product, which can then be transferred to one of the above-mentioned units and thus used in the optimization method according to the invention.

[0053] In addition, particularly advantageous designs and improvements of the present invention are derived from the dependent claims and the following description, wherein independent claims in one category of claims may also be further formed by analogy into dependent claims and embodiments in another category of claims, and in particular individual features of different embodiments or variants may also be combined to form new embodiments or variants.

[0054] As described above, in the optimization method, at least one optimal parameter set may also be determined as at least one further optimized process variable value (taking into account the requirement data). Such an optimal parameter set may preferably be selected from a number of "candidate parameter sets", wherein this selection is preferably also performed using a KI-based optimization unit. Here, this may be a separate KI-based optimization unit, the sole purpose of which is to find an optimal parameter set (e.g. a neural network trained specifically for this purpose). It may also be a KI-based combined optimization unit (e.g. constructed using the above method), which may be used to find multiple optimized or optimal process variable values ​​at the same time, for example to find an optimal pairing of a scan direction distribution and a parameter set.

[0055] As already mentioned above, the parameter set (also synonymously referred to as "process parameter set") and therefore also the candidate parameter set each comprises a defined tuple of individual process parameter values, which are then used to control or optimize the control of the machine to construct at least one layer of the relevant segment. The process parameter set can in particular be a predefined, preferably discrete (ie non-continuous) optimization variable.

[0056] The parameter set preferably includes one or more of the following process parameters:

[0057] - The power of the energy beam (e.g. the laser power in laser melting process)

[0058] -Scanning speed of energy beam

[0059] - Fill distance

[0060] -Energy beam diameter (e.g. focus spread and shape)

[0061] -Intensity distribution or intensity profile of the energy beam

[0062] - When laser is used as energy beam: continuous or pulsed working mode

[0063] -Energy beam power curve

[0064] -Thickness of the layer

[0065] For different types of building materials, for example different powder types, preferably metal powders, a plurality of candidate parameter sets can be provided each time. The different powder types can be distinguished in particular according to the following criteria:

[0066] a) Materials, where a distinction is made between pure materials or alloys,

[0067] b) Other powder parameters, such as particle size distribution, particle sphericity, chemical properties, etc.

[0068] Since different powder batches of the same material may have different combinations of the above parameters, each powder batch may also be considered as a separate powder type if necessary and appropriate.

[0069] However, each parameter set or candidate parameter set may also include the relevant construction material type itself as another "process parameter value", that is, when a candidate parameter set is selected, the material type is determined by the process parameter value (discrete value). This is ultimately a matter of organizational or structural design of the candidate parameter set database.

[0070] In practice, initially only a few candidate parameter sets may be used, for example 4 to 20 candidate parameter sets may be used for a particular material. In principle, the number of candidate parameter sets is limited only by the technical possibilities of the database size, i.e. how much storage space and how much computing time are available (in advance) for creating the database. When determining the number of candidate parameter sets, the required computing time may also be taken into account, since limiting the number reduces the computing time in the optimization method. The main advantage of using candidate parameter sets is that computing power can be saved.

[0071] As mentioned above, preferably (e.g. at the beginning or before the optimization method), the product is virtually divided into a plurality of segments using the requirement data, in particular the geometric data, and then the optimization method is performed to determine the optimized process variable values ​​for each segment, preferably the optimized segment scanning direction distribution and the optimal parameter set for each segment. The determined optimal segment scanning direction distribution can then jointly form the optimal scanning direction distribution of the entire component.

[0072] For this purpose, first a so-called "domain" (which may also be called "computational domain" or "design space") is preferably defined, which includes the product, ie the product is completely contained within the domain.

[0073] This entire domain is then (virtually) divided into so-called "segments", wherein a product comprises at least one such "segment". In general, a segment may be a region in a domain, typically a component.

[0074] However, the domain may also include so-called "powder segments", i.e. segments that are not solidified or to be solidified. This may be, on the one hand, areas within the "domain", but outside the contour of the component (but within the construction space or within the production volume of the additive construction machine or within the design space), and on the other hand, cavities or voids in the product. As will be described later, the final contour of the component can only be defined by the boundary between the segments to be solidified and the powder segments.

[0075] If the domain is, for example, a cuboid surrounding the raw product, which is at a certain distance from all sides of the cuboid, then in the simplest case, two segments are sufficient within the domain, namely a solidification segment or a segment to be solidified (which includes the entire raw product) and a powder segment (which includes the entire area outside the raw product). In principle, however, the domain boundary can also completely coincide with the raw product boundary or contour, wherein, for example, there need not be any powder segment at all unless there are cavities in the raw product within the domain.

[0076] This segmentation of components or entire domains can be performed automatically or according to user requirements with the help of a user interface, wherein a semi-automatic method is also possible, i.e. partly automatically and partly according to user requirements. The segmentation preferably uses requirement data, in particular geometric data. For example, the components can also be subdivided according to certain functionally necessary structural parts (i.e. the functions mainly performed by the structural parts), such as into struts, pressure plates, flange parts, etc.

[0077] In a particularly preferred embodiment of the method according to the invention, the optimization method is performed using a defined "objective function", suitable and preferred objective functions are also described in detail in DE 10 2022 117 935, for example, to which reference is therefore again explicitly made here. For example, the objective function and the requirement data can be used to select at least one best "parameter set" from a number of "candidate parameter sets" for at least one segment of the product within a defined domain and to determine an optimized or optimal segment scanning direction distribution (ultimately matching the best parameter set).

[0078] In the optimization method, each objective function assigned to the relevant segment can preferably be selected in such a way that, if necessary, certain boundary conditions (e.g., maximum permissible Mises equivalent stress or minimum safety factor for a given external load) are achieved as well as possible in the segment to achieve predetermined target macroscopic properties (e.g., quality requirement data, in particular load data regarding the loads to which the component has to withstand, such as high stiffness at the highest possible construction rate, while complying with a certain determined safety factor of, for example, 1.65), provided that the optimal process variable values ​​obtained by minimizing the objective function (or at least a sub-function matched to the target macroscopic property) are maintained as well as possible or are as close as possible to each other in the subsequent additive construction process. In addition to the geometric data, quality requirement data can also be part of the above-mentioned requirement data.

[0079] It is required that data optimization can also be taken into account (directly or indirectly), at least partially, in defining the objective function.

[0080] Likewise, some of the required data, in particular geometric data, can also be taken into account in the domain definition. For example, certain conditions can be set via the external shape of the domain, for example by fitting the product into the domain and extending, for example, to a certain external surface of the domain. The boundary condition in the objective function can then be that the material must solidify in a certain area of ​​the domain.

[0081] In the optimization method, it is particularly preferred to select one optimal parameter set from the candidate parameter sets for a segment, i.e. for all layers in the segment, because this is much less computationally intensive than searching for a plurality of optimal parameter sets assigned to different layers of the segment. Likewise, each segment preferably has only one optimized segment scanning direction distribution. In other words, a segment can very preferably also be defined in such a way that there is exactly one optimal parameter set and one optimized segment scanning direction distribution within the segment boundary. The optimal parameter set and / or the optimized segment scanning direction distribution then changes at the boundary from one segment to another.

[0082] Particularly preferably, in the optimization method, the optimized process variable values ​​for a plurality of segments within the definition domain can also be determined in parallel (ie coupledly) using a common objective function.

[0083] In this case, all segments of a component or even all segments of a domain are very preferably coupled in one optimization method. When there are conflicting goals between the requirements of individual segments or of the entire component, the solution of the optimization method (i.e. the optimal set of parameters obtained by the optimal segment scanning direction distribution for the segment) can also be the Pareto optimal solution for the entire product. Therefore, an optimized scanning direction distribution (i.e. component scanning direction distribution) will also be found and determined for the entire component.

[0084] Determining the optimized process variable values ​​of multiple segments in parallel (i.e. simultaneously) by means of a "common objective function" can also be understood as using several mathematically coupled piecewise objective functions, wherein a single piecewise objective function is respectively assigned to one of the segments. This coupling can ultimately be used to determine the optimal parameter set and the associated optimized segment scanning direction distribution with a common objective function (defined by the piecewise objective function) for the entire domain, i.e. all segments defined therein. In other words, the common objective function is actually the sum of the piecewise objective functions of all segments involved in the joint optimization.

[0085] The objective function very preferably includes as further requirement data the minimization of the variation of the parameter set within the entire product. Considering that this additional objective is equivalent to reducing the segment boundaries as much as possible, the product is divided into as few (virtual) segments as possible. In other words, this can also be achieved by formulating the objective function in such a way that the segment boundaries are minimized.

[0086] In particular, the segment boundaries are preferably taken into account as further optimization variables in the optimization method and are then provided as further optimization process variable values ​​at the end of the optimization method, i.e. after the optimization is completed. In other words, the segment boundaries can also be moved within the scope of the optimization. In extreme cases, the change of the segment boundaries can even make the segments disappear completely. New segments can also be created by moving the segment boundaries. Therefore, in this preferred variant, the number of segments within a domain is not necessarily fixed, but can also be optimized in the optimization method. In particular, the number of domains can also be minimized in this way in order to achieve the goal of modifying the optimal parameter set as little as possible in the component.

[0087] In particular, the movement of the segment boundaries also affects the outer boundaries of the raw product, since these boundaries are defined as segment boundaries between component segments and powder segments in the domain. In this way, the topology of the component can also be modified in the optimization method, i.e. the shape of certain regions can be different from the shape initially specified in the starting specification, for example, if the modified topology can better realize the component requirements, or at least can fully realize the requirements with less effort. In this regard, the geometric data of the raw product originally specified as requirement data can also be changed or optimized, in particular if it can more accurately define the shape of the raw product.

[0088] As a result of further development of the optimization method, the following optimized process variable values ​​were obtained for segment optimization:

[0089] 1. The best parameter set (as the first optimized process variable value), which in turn comprises a tuple of individual process parameter values,

[0090] 2. Optimize the segment scanning direction distribution (as the second optimization process variable value).

[0091] 3. Optimize segment boundaries (as the third optimization process variable value).

[0092] In order to realize the movement of the segment boundaries in the optimization method, phase field methods, in particular multiphase field methods, can preferably be used, as described in patent applications DE 10 2022 117 935 and DE 10 2022 117 936. Therefore, reference is also made to these documents, in particular with regard to the phase field method. Multiphase field methods are particularly suitable for processing movable segment boundaries.

[0093] It is particularly preferred that the parameter sets assigned to the segments can be distributed proportionally at least at the location of the "interface region" between several adjacent segments (at least two different adjacent segments, possibly more than two adjacent segments).

[0094] In the optimization method, the parameter sets present at a location are preferably also each represented by a "ratio" of the parameter sets. The ratio value is preferably between 0 and 1, where a parameter set ratio value of 1 indicates that the parameter set is present at the location and a ratio value of 0 indicates that it is not present. Thus, in the optimization method, a location in the interface region between two segments can be simply characterized by a first parameter set ratio applicable to the first segment and a second parameter set ratio applicable to the second segment. In the interface region where more than two segments meet, a location can also have more than two parameter set ratios. The sum of all parameter set ratios at each location is preferably equal to 1.

[0095] In the optimization method, the width of the "interface region" (which is then generally assumed in the method) may preferably be defined or specified by the user.

[0096] In the optimization method, one of the process parameter values ​​in the (optimal) parameter set for a single layer of the segment comprises at least one layer scanning direction arrangement, i.e. a scanning direction which is to be assigned or has been assigned at each time in the respective layer during the construction process. In particular, the layer scanning direction arrangement may comprise a filling direction arrangement (filling strategy) in the layer. Thus, as described above, in each layer there is a "scanning direction distribution within the layer" which is determined by the layer scanning direction arrangement.

[0097] It is particularly preferred to select in the optimization method a layer scanning direction arrangement that can be applied to all layers of a segment, except for a possible rotation of the overall orientation of the layer scanning direction arrangement between different layers. The segment scanning direction distribution is then generated as a rotational combination of the layer scanning direction arrangements between the layers in the segment. To optimize the segment scanning direction distribution, the relative orientations of the layer scanning direction arrangements of the different layers of the segment can then preferably simply be optimized with respect to one another, wherein the rotations of the layer scanning direction arrangements between the layers in the segment can be defined by appropriate control commands, which can be used to control the production device when constructing the component. In particular, it is convenient for these optimization (sub)problems to use a KI-based optimization unit.

[0098] The at least one process variable value in the parameter set preferably also includes a track width between two reinforcement paths, ie, for example, a selected filling distance. The track width can be set in the parameter set independently of the slice scanning direction arrangement.

[0099] Preferably, in the optimization method, for example in the objective function, the alignment of the product based on the main construction direction, i.e. the relative orientation in the construction space, is taken into account as a further optimization variable. In the case of layer-by-layer construction, the main construction direction is usually considered to be perpendicular to the layer direction, in which the layers are gradually deposited. In beam melting processes, especially laser melting processes, a Cartesian coordinate system x, y, z is usually defined as a reference system, wherein the x- and y-directions are parallel to the layer or across the plane of the construction field, and the z-direction is perpendicular from the construction field upwards, i.e. coincides with the main construction direction.

[0100] At the end of the optimization method, i.e. after the optimization is completed, the found optimal orientation can be used as a process variable value for further optimization. This is advantageous because the orientation in the build space influences the position of the segment boundaries in the space. By taking the orientation into account, it is also possible to achieve, for example, a reduction or even minimization of overhangs and / or supporting structures (support structures) during the optimization.

[0101] In the optimization method, different requirement data may be taken into account, for example, in an objective function or in other ways. The requirement data may preferably include one or more "target production data" and / or "target property data" and / or "constraint conditions".

[0102] Particularly preferably, one or more of the following target production data can be taken into account:

[0103] -Building rate during additive construction,

[0104] - the material type of the building material (with which not only the material but also the concentration can be specified, e.g. whether it is a powder and, if so, with which parameters),

[0105] - construction technology (i.e. type of construction method, e.g. laser melting, electron beam melting, etc.),

[0106] - Machine type (i.e. the type of production equipment used).

[0107] It is also preferred that one or more of the following target attribute data be considered:

[0108] - Target load data (which may, for example, include information about the external loads that the component has to withstand. However, these external load information may also already include stress states determined by these external loads (e.g. within the scope of the simulation), i.e. "internal loads" in the corresponding region of the component and / or in the entire component.)

[0109] - Stiffness (i.e. the elastic deformation resistance of the product in each section)

[0110] - Strength (i.e. the resistance of the product to plastic deformation in each section.)

[0111] - the mass and / or mass distribution of the product (in many cases, the aim here is to achieve the lowest possible mass, i.e. to reduce the mass, so that the product is as light as possible and / or to save material costs. However, depending on the component, it may also be intentional to have the highest possible mass at least locally, e.g. flywheel mass or the like)

[0112] - Surface accessibility (For example, certain requirement data about the accessibility of the surface can ensure that post-processing of the component is guaranteed or becomes easier. For example, good removability of support structures generally requires good accessibility. It is also preferred to check the accessibility in a separate method step or process step in the optimization method by suitable methods, such as ray tracing or as described in M. Inui, S. Nagano and N. Umezu, Rapid calculation of the reachable cone for assisted 3+2-axis milling; Computer-Aided Design and Applications, 2018, Vol. 15, No. 5, pp. 667-676.)

[0113] -Support properties (here, the properties that the support structure should have can be considered, such as whether it is used for support and / or heat dissipation. It should be noted that the support structure used for heat dissipation does not necessarily have to be firmly connected to the component; there can also be a powder layer between the support structure and the component, and the powder layer corresponds to at least one (actual) layer thickness (in addition, the overhanging part of the component can also be designed in this way, that is, it can be used without a "support frame").

[0114] Preferably, one or more of the following constraints should also be considered:

[0115] - chemical properties (e.g. component material should not rust),

[0116] - Geometric data (e.g. specific dimensions to be strictly adhered to or maximum / minimum dimensions of a component, as mentioned at the beginning).

[0117] In addition, various other requirement data may be considered depending on the type of product (component). Incidentally, some requirement data may be considered as both "target production data" and "target attribute data" or "constraints". Similarly, as mentioned above, some data, particularly target attribute data related to component load-bearing capacity or chemical properties or chemical resistance, may also be considered as quality requirement data.

[0118] Particularly preferably, the requirement data are taken into account in the optimization method by means of predefinable weightings, ie it is possible to set which requirement data, for example which requirement data are more important and which are relatively less important.

[0119] As in DE 10 2022 117 935, the objective function preferably consists of a plurality of sub-functions, each of which is assigned specific requirement data, i.e. each sub-function represents a specific requirement. The requirement data are therefore taken into account in the optimization method by means of predeterminable weights, which can also be achieved here by the objective function comprising a sum of weighted sub-functions, wherein the sub-functions are assigned to specific requirement data.

[0120] Preferably, within the scope of the present invention, an optimization method comprising a plurality of iterative steps is used. In an iterative optimization method, in particular, individual sub-functions can be optimized independently of other sub-functions or optimization parameters in separate iterative cycles. Depending on the specific design, this can reduce the amount of calculations.

[0121] Particularly preferably, at least one KI-based optimization unit is used in at least one iteration step of the iterative optimization method.

[0122] Therefore, the entire optimization method can preferably also be used as a "hybrid method", i.e., a mixture of KI-based and "classical" (not KI-based) optimization schemes, wherein individual classical optimization steps (which may also further include iterative loops) can be replaced by KI-based optimization units.

[0123] Preferably, in the optimization method, at least one starting process variable value (e.g., a starting parameter set and / or a starting scanning direction distribution, in particular a starting segment scanning direction distribution) is first set. Preferably, a "starting configuration" is determined, which includes a combination of a plurality of starting process variable values. For example, to determine the starting configuration, at least a starting segment can be defined or set, and a starting parameter set can be selected for each starting segment from a number of candidate parameter sets, while the starting segment scanning direction distribution is determined. For example, the starting configuration can be selected in a first step of the optimization method immediately after defining the domain.

[0124] In a preferred variant, the candidate parameter set with the highest construction rate in the segment is selected as the starting parameter set of the segment. However, the starting parameter set can also be selected in a different way, such as a simple random selection.

[0125] Particularly preferably, in the optimization method, at least one initial process variable value can first be determined using a KI-based optimization unit. That is, the KI-based optimization unit(s) select, for example, appropriate starting values ​​or "pre-optimization values" so that subsequent further optimization methods (e.g., mainly classical optimization methods) can "converge" faster, i.e. reach the goal faster. In particular, at least one KI-based optimization unit can be used to determine a starting configuration (as described above, at least defining a starting segment and selecting a starting parameter set from several candidate parameter sets for each starting segment, and determining a starting segment scanning direction distribution).

[0126] It should be pointed out here that for the above-mentioned "powder segments" (i.e., segments in the domain that do not need to be solidified), for example, the energy beam or laser power can be simply set to 0 in the initial parameter set, that is, no energy is introduced into these powder segments. This value will remain permanently in the powder segment, that is, it will not be changed in the optimization method or iterative method. On the other hand, if the component topology needs to be optimized in the optimization method, the powder segment boundary can of course be transferred to the adjacent powder segment.

[0127] In principle, various criteria and / or methods may be considered for determining the optimized scanning direction distribution and / or for selecting the optimized or best parameter set from the available candidate parameter sets.

[0128] In a preferred procedure, at least one parameter set fitness value is determined for at least one area of ​​the product (e.g., a segment; i.e., by segment), for at least several possible (candidate) scanning direction distributions (or candidate segment scanning direction distributions) and / or for at least a portion of the candidate parameter sets.

[0129] The parameter set fitness value may be a scalar value, preferably between 0 and 1, which represents a measure of the fitness of the associated candidate parameter set to satisfy the specific requirement data. It is also referred to below as the "parameter set score" (or "PS score" for short). Reference is made in particular to patent application DE 10 2022 117 935 in this regard.

[0130] For example, the PS score of a candidate parameter set can be used, for example, to determine whether the particular candidate parameter set (under a particular scan direction distribution) is the best candidate parameter set for satisfying certain defined requirement data, compared to the PS scores of other possible candidate parameter sets, or the PS score can be viewed as a probability measure of the candidate parameter set (under a particular scan direction distribution) being the best parameter set. For example, a candidate parameter set with a PS score of almost 1 may also be almost 100% suitable for satisfying the requirement.

[0131] Then, the parameter set fitness value is used to determine an optimized scanning direction distribution (or a segment scanning direction distribution for a certain segment) and / or to select the best parameter set from the candidate parameter sets.

[0132] Here, it is particularly preferred that parameter set fitness values ​​are determined for different pairings of segment scanning direction distributions and candidate parameter sets. That is, for each segment for which an optimal parameter set and optimized segment scanning direction distribution need to be found, parameter set fitness values ​​are calculated, and these parameter set fitness values ​​are optimized. Here, for example, a pairing of a parameter set and a segment scanning direction distribution may be selected for a certain segment to obtain an optimal parameter set fitness value.

[0133] Preferably, multiple requirement-specific parameter set fitness values ​​(i.e. requirement-specific PS scores) for different requirement data are determined for at least a part (segment) of the scanning direction distribution and / or the candidate parameter set (especially each pair of parameter set and segment scanning direction distribution).

[0134] That is, the requirement-specific PS score can be used as a comparison metric to clarify which of the available candidate parameter sets (in combination with a specific scan direction distribution or segment scan direction distribution, if necessary) are best suited to satisfy precisely defined specific requirement data, such as a required build rate and / or intensity. An example of how possible (requirement-specific) PS scores can be determined will be provided later.

[0135] Particularly preferably, the requirement-specific parameter set suitability values ​​of (in particular each pair of) scanning direction distributions and / or candidate parameter sets can each be combined to form an overall parameter set suitability value (for the respective segment).

[0136] Since in optimization methods it is often necessary to select the best parameter set and / or (segment) scanning direction distribution among a plurality of different, partially even conflicting requirements, it makes sense to use such an overall parameter set fitness value. The overall parameter set fitness value of the candidate parameter sets can then be used to select the best parameter set from the candidate parameter sets.

[0137] Examples of suitable combinations of possible (requirement-specific) PS scores are given below, which are also described in DE 10 2022 117 935. The type of combination can also depend on the requirements.

[0138] The combination method may preferably include the product of the requirement-specific parameter set fitness values. In particular, the overall parameter set fitness value may be obtained by simply multiplying all requirement-specific parameter set fitness values ​​of the relevant candidate parameter sets.

[0139] Very preferably, in the optimization method, optimized process variable values ​​for the product can be determined in such a way that optimized process variable values ​​are determined for each segment (preferably an optimal parameter set and an optimized segment scanning direction distribution are determined respectively), which are optimized on the one hand with respect to the total parameter set fitness value in the respective segment and with respect to the "sum of parameter set fitness values" in the product. For example, such a sum of parameter set fitness values ​​can be formed by adding the total parameter set fitness values ​​of the individual segments over all segments of the product.

[0140] Therefore, in this preferred procedure, all segments are also optimized in parallel using a common objective function, wherein preferably the sum of the parameter set fitness values ​​is used as at least one main part of the objective function, eg to be maximized.

[0141] Here, within the scope of the present invention, a KI-based optimization unit is preferably used to determine the segment scanning direction distribution and / or select the best parameter set from the candidate parameter sets for a certain segment (especially select a pairing of the scanning direction distribution and the candidate parameter set).

[0142] Preferably, at least one KI-based optimization unit is used, during the generation of which the KI-based optimization unit (in particular a neural network) is trained using parameter set fitness values ​​(in particular using total parameter set fitness values).

[0143] The selection of scan direction distribution and candidate parameter set pairings to obtain the "best" (e.g., largest) overall parameter set fitness value can be significantly accelerated if a KI-based optimization unit (e.g., a neural network) trained for the best overall parameter set fitness value is used at segment level.

[0144] That is, by training a KI-based optimization unit using parameter set fitness values ​​(especially requiring specific parameter set fitness values) and / or total parameter set fitness values ​​and / or the sum of parameter set fitness values, and subsequently using at least one trained KI-based optimization unit, the optimization process itself is ultimately made faster (although almost "indirectly") also by virtue of the relevant fitness values ​​(only generally speaking).

[0145] Here, it is preferred that the process variable values ​​that are optimized overall for the sum of the parameter set fitness values ​​in the product are determined by using an (iterative) combinatorial optimization method, particularly preferably a heuristic approximation method, more particularly preferably a simulated annealing method and / or a quantum annealing method.

[0146] Simulated annealing methods or quantum annealing methods may be particularly suitable for finding approximate solutions to optimization problems which, due to their high complexity, preclude the exhaustive attempt of all possibilities and mathematical optimization methods.

[0147] Therefore, within the scope of the present invention, such methods are particularly suitable, even if many possibilities need to be tried, because various variants can be determined relatively quickly and "computationally cheaply" by means of KI-based optimization units (such as neural networks). Thus, simulated annealing methods can be used in conjunction with KI-based optimization methods, for example, and are particularly useful here for solving combinatorial problems with parameter swaps.

[0148] As mentioned above, the optimization method preferably comprises a plurality of iterative steps, i.e. at least a part of the method can be iteratively run a plurality of times. Here, for example, in one or more steps, the scanning direction distribution can be (pre-)determined (if necessary starting) and the (if necessary starting) optimal parameter set can be selected, for example using the (total) PS score, in one or more other steps, the optimized segment scanning direction distribution and the optimized segment boundaries can be determined, for example using an objective function or a sub-function, and if necessary, further optimized process variable values ​​can be determined in other steps (with or without an objective function), wherein changes to the scanning direction distribution and the parameter set may also be made, as explained later by way of example. In all these steps, as mentioned above, KI-based optimization units can also be useful, for example, for completing sub-steps in "classical" optimization steps, if necessary, or for training KI-based optimization units using "classical" optimization steps (or using the results of these optimization steps as training data) so that they can later replace these optimization steps.

[0149] An iterative loop consisting of a plurality of steps can be run a plurality of times until a predetermined termination criterion is reached. This termination criterion is preferably met if the value of the process variable found in the current iterative loop is the best value, i.e. no significantly better value is found in the new run, and / or all requirements are fully met according to a predetermined evaluation criterion, and / or, for example, a certain number of runs is reached, etc. Other termination criteria are also conceivable.

[0150] The optimization method preferably includes at least one state determination step, in which a "state description" of a product consisting of the desired construction material is determined using the current process variable values. Thus, in an iterative process, the "current process variable values" are the process variable values ​​applicable to the current iteration of the iterative cycle. In a first run, the current process variable values ​​are the process variable values ​​of the above-mentioned starting configuration.

[0151] In order to determine the state description in the state determination step, the state of the current system, i.e. the state of the component with the current segments and the parameter sets currently assigned to the segments, is preferably simulated (i.e. the relevant segments of the virtual product for which the optimal process variable values ​​are currently being sought will behave, for example, under a certain load, as if it were produced using the current process variable values). Therefore, the state determination step can also be called a "state simulation step". Particularly preferred simulation methods include, for example, the finite element method or finite volume simulation. For example, a load simulation or a vibration simulation can be performed on the (virtual) component, the result of which is the possible load or natural frequency of the system or component under the assumption of the current configuration of the process variable values. In particular, the expected stress states in the component can be determined here, which can be used as input data, for example, in a KI-based optimization unit (especially a neural network) in order to obtain an optimized scanning direction distribution or an optimal parameter set according to these load requirements.

[0152] The state description is preferably compared with predetermined quality requirements for the raw product. This can be used to check whether the raw product meets the predetermined quality requirements. The state simulation step can be performed as a (quality) requirement simulation, i.e. using quality requirement data that specifies how the component may or should behave under a certain load or a certain force. In particular, the state simulation step can use at least part of the requirement data, which can also include appropriate quality requirement data. The requirement data can thus be used when selecting the best parameter set and the objective function.

[0153] If the state description does not meet the predetermined quality requirements, then the current process variable value is preferably (further) changed. This further change can be performed in a further separate optimization process step or method step, as will be described later, or can be integrated into the various steps of the subsequent method.

[0154] Optionally, after a further change in the value of the process variable, the state determination step can be performed again and the state description can be compared with the predetermined requirements. In other words, this check can also be performed in an iterative loop. For example, the termination criterion of the iterative loop can be success (the state description meets the predetermined requirements) or reaching a maximum number of iterations. If necessary, it is possible to start again by changing the starting configuration (e.g. also using different materials).

[0155] The optimization method can then include various further optimization process steps (e.g., also in separate iteration cycles), as described, for example, in DE 10 2022 117 935. In principle, all of the above-described steps can also be advantageously used here (wherein, if advantageous, these steps can be supported or replaced by the use of a KI-based optimization unit, if necessary).

[0156] At the end of a series of iterative method steps, an improved segment with an improved current parameter set and an improved segment scanning direction distribution is preferably present, ie an improved configuration or an improved current process variable value is then present.

[0157] In a particularly preferred further development of the method according to the invention, a property database of a property database system is used within the scope of the optimization method (e.g. in one of the above-mentioned steps) to determine or select a changed (updated) segment scanning direction distribution of the segments. In such a property database system, the properties of the raw product to be constructed, or more precisely the properties of the individual layers of the raw product and / or the segments formed therefrom, can be stored as a function of the relevant process parameter set of the relevant layer or the relevant segment and, if applicable, also as a segment scanning direction distribution function.

[0158] There are many possibilities for realizing such an attribute database system. In particular, the attribute database system can also be composed of a plurality of attribute databases, for example with different attributes and / or parameter assignments.

[0159] The property database system preferably comprises a so-called "basic property database". The "basic properties" of individual layers can be stored in this basic property database, depending on the process parameter set to be or already used for constructing the layer (including the layer scanning direction arrangement or the filling direction arrangement or the type of construction material, which are also process parameters of the respective process parameter set). Therefore, in such a database, each parameter set is assigned at least one basic property value, preferably a set of basic property values, which the layer of the segment or component will have if the corresponding layer is constructed using the assigned parameter set.

[0160] Methods for constructing and using such a basic property database are described in detail in DE 10 2022 117 935 and in particular also in DE 10 2022 117 936, the contents of which are hereby fully incorporated. As described in these documents, in suitable test methods, previously produced test specimens can be used to determine at least one basic property value and / or microstructure for each of these test specimens, which basic property values ​​and microstructures can be stored or saved as entries in the basic property database in association with a parameter set for producing the test specimens, which parameter set can preferably include, in particular, the type of construction material and the layer scanning direction arrangement or the filling direction arrangement / filling strategy.

[0161] In particular, macroscopic properties or “macroscopic property values” of a segment or even an entire component formed by the layer can be determined from these basic properties of the layer.

[0162] This "macro-property value" describes the property value at the macro level or macro perspective, that is, the property possessed by the entire segment, such as thermal conductivity, fracture strength, etc. In this method, it is preferred to simultaneously determine the macro-property values ​​of a component or multiple segments of a component. The macro-property value may include tensor values, such as elasticity tensor, and may also include classification values, such as corrosion resistance or not, the nature of the lattice structure, such as face-centered cubic (kfz), body-centered cubic (krz) or hexagonal close packing (hdp). Various macro-property values ​​will be described later.

[0163] If the properties of the individual segments of a component are known at a macro level, i.e., the "macro-property values", this can also provide information about the properties of the component and the overall quality of the component, in particular whether the component meets certain quality requirements. Therefore, the macro-property values ​​of the segments can also be used in the above-mentioned state determination step to determine the state description of the product.

[0164] The basic property databases for a plurality of different parameter sets can each contain as basic property values ​​a layer "texture" which was generated during the additive construction process using the respective parameter set (i.e. also using a specific construction material). The term "texture" refers to the sum of the crystal orientations in the structure, i.e. the crystallographic texture, which should not be confused with surface textures, such as surface roughness. The texture is particularly preferably described in the form of a so-called "orientation density distribution function" (ODF). The texture or ODF can be determined, for example, by measurements using the EBSD method (EBSD=electron backscatter diffraction) in a scanning electron microscope or other methods.

[0165] Alternatively or particularly preferably in addition, the basic property database may also include further basic property values, for example, which can also be determined based on the texture of the layer of the parameter set, in particular the orientation density distribution function. Further basic properties can be calculated from the texture or ODF using the known properties of the single crystal of the construction material (for example by averaging or homogenization methods, as also described in detail in DE 102022 117 936). For example, such basic properties can be the yield point, the tensile strength in any direction, etc., to name just a few. Conversely, the texture can also be derived from other basic property values ​​or macroscopic property values, such as the elasticity tensor.

[0166] The basic property database preferably includes basic property values ​​of reference orientations of the scanning direction arrangement of each layer, especially the filling direction arrangement. The reference orientation or reference arrangement can be selected arbitrarily.

[0167] For layers whose layer scanning direction arrangement and their "scanning direction distribution within a layer" are rotated by at least one rotation angle (in any direction about the main construction direction, i.e. about the direction perpendicular to the layer) compared to the reference orientation, the basic property values ​​can be determined or calculated from the corresponding basic property values ​​stored for the reference orientation using the rotation angle each time. This can be achieved by a simple angle conversion. Rotation of the layer scanning direction arrangement, in particular the filling direction arrangement, from layer to layer is very common, for example in the jet melting process. For example, the rotation angle from layer to layer is usually 67°.

[0168] There are several ways to determine the value of a segment's macro attributes.

[0169] As described above, in a preferred method, the macroscopic property values ​​of a segment having a plurality of superimposed layers are determined or combined from the basic property values ​​of the individual layers. This is preferably achieved using a mathematical "homogenization method". As mentioned above, a corresponding method is explained in detail in DE 10 2022 117936, so reference can be made to this method.

[0170] In order to further save computing time, for example in configurations that occur repeatedly within a segment, preferably, as described above, at least one macroscopic property value of at least one segment can be determined using the provided basic property database. Therefore, alternatively or additionally, the property database system preferably also includes a so-called "macroscopic property database". In this database, at least one macroscopic property value can be stored per segment (consisting of several layers), preferably a set of macroscopic property values, each for a different combination of segment scanning direction distribution and parameter set (also depending on the construction material), which macroscopic property value is to be or has been created by the segment scanning direction distribution and parameter set assigned in the database.

[0171] When determining or selecting the segment scanning direction distribution after a segment change, it is preferred to consider whether macro-property values ​​have been entered in the macro-property database for possible segment scanning direction distributions and (for example, in an optimization method) specific combinations of the current parameter set (including construction materials) that may be predetermined in the next step (ie, "candidate combinations").

[0172] If this is the case, it can be decided whether to use a stored segment scanning direction distribution (and therefore in particular a filling direction arrangement in a single layer or a "standard" filling strategy) for the segment to be constructed, which may be more advantageous in terms of computing technology and time, but the construction may be slower, or whether to use a not yet stored strategy with a single filling direction arrangement, which may be faster and / or have other advantages, but requires more complex calculations based on a single basic property value.

[0173] However, if the "standard" construction strategy and especially the "standard" filling strategy cannot be used, more complex calculations based on basic property values ​​must be performed anyway.

[0174] On the one hand, it is much simpler and faster to determine the macro-attribute value of a complete segment by querying the macro-attribute database than to determine the macro-attribute value of a segment based on the basic attributes of a single layer. On the other hand, creating and storing a large number of macro-attribute values ​​requires a lot of computing time and storage space.

[0175] Therefore, the macro-property database preferably contains at least macro-property values, preferably sets of macro-property values, for the most commonly used construction strategies, in particular the "standard exposure strategy" or the so-called "standard filling strategy" often used in beam melting processes. Typical standard filling strategies in beam melting processes are so-called 67° filling or xy filling (= 90° filling). In these methods, the orientation of the filling strategy is rotated by 67° or 90° from layer to layer, while the filling strategy remains essentially unchanged.

[0176] If certain queries occur many times, it makes sense to include them in the "standard" filling strategy entries of the macro-attribute database. The database system is therefore preferably designed in such a way that it can record which combinations of segment scanning direction distributions and parameter sets are used particularly frequently and then create new entries in the macro-attribute database accordingly, that is, the database system can be said to "learn".

[0177] As mentioned previously, in addition to texture or ODF, there are other property values ​​(especially fundamental or macroscopic property values) that may be of interest. These property values ​​can usually be calculated from the texture or ODF (e.g., by averaging) using known properties of a single crystal of the construction material.

[0178] Particularly preferably, the at least one property value, in particular the basic or macroscopic property value, comprises at least one value of one of the following material parameters:

[0179] - Elasticity tensor

[0180] - "tensile strength tensor" (represents the mechanical stress at a certain location in the workpiece when a specific yield criterion exists; the definition of the tensor variable entries for each yield criterion can be found in J. Betten, Kontinuumsmechanik, 1993, Springer)

[0181] - Yield point distribution (e.g. in the form of Hill tensors, also found in the book by J. Betten)

[0182] -Curing coefficient

[0183] - Thermal conductivity

[0184] - Breaking strength.

[0185] Preferably, the property value of at least one material parameter may include a plurality of direction-dependent fractions, i.e. the property value may also be anisotropic. In general, the property value may be defined as a tensor, such as a vector (first-order tensor) or a matrix (second-order tensor), in order to take into account three dimensions or directions, or as a fourth-order tensor, in order to take into account properties in a crystal system.

[0186] An example of this is the fourth-order elasticity tensor, where the entries of the elasticity tensor for different crystal space directions contain values ​​of the general three-dimensional stress state from which the elastic modulus E can be calculated by transformation, for example in the x-direction and the y-direction in the layer.

[0187] Similar anisotropies can also appear in the yield point distribution or the tensile strength tensor. Without limiting the generality, other common visualization forms such as the Voigt symbol can also be used.

[0188] The optimization method preferably includes at least one "cavity test step". This step can be used to test whether any cavities present in the raw product after construction, which may be filled with uncured powder, are connected to the surface of the raw product. The purpose of this is to check whether the powder can be removed from the component cavity (and if so, how effectively it can be removed). The cavity test step can therefore also be called a "powder removal test step". For the detailed procedure, reference can be made again to DE 10 2022 117 935.

[0189] If it is found in the cavity testing step that not all cavities can be depowdered as expected, the component geometry can be corrected again if necessary. For example, the optimization method can be started from scratch, in particular with a different set of starting parameters.

[0190] In addition, the optimization method preferably also includes at least one heat conduction test step, in which it is checked whether the product can be subjected to a given heat treatment in accordance with a predetermined quality standard, that is, the heat treatment requirements of the component are checked. In particular, it is checked whether the heat treatment can be carried out within a reasonable time and whether a sufficient final quality is guaranteed. If not, the optimization method can also start from scratch, in particular using a different starting parameter set. However, in a particularly preferred variant of the present invention, the heat treatment requirements can also be taken into account when determining the optimized scanning direction distribution or the optimal parameter set in the KI-based optimization unit (in particular a neural network), in that the time-temperature curve that meets the load requirements is passed as input data to the KI-based optimization unit, as explained later by way of example.

[0191] As described above, control data for a production device for additive building of a product may then be generated based on the optimized process variable values ​​so that the optimized process variable values ​​are adequately achieved in a layer-by-layer additive building process according to specified evaluation criteria.

[0192] Preferably, an optimal orientation of the slice scanning direction arrangement, i.e., in particular the orientation of the filling direction arrangement or the filling strategy of the individual layers, is selected for the individual layers in the segment in order to achieve or approximate the best possible optimal segment scanning direction distribution over all layers in the segment. In other words, the initially continuous optimization variable "scanning direction distribution" (in particular "segment scanning direction distribution") is discretized based on the control parameter in order to take into account in the slice-by-slice construction that only one predetermined slice scanning direction arrangement or filling strategy exists in each slice, more precisely, preferably the slice scanning direction arrangements are identical but only rotated relative to each other.

[0193] As described above, the above method allows for an overall optimization of the property profile of an additively constructed part. The method takes into account the correlation between the selected manufacturing strategy, in particular the selected manufacturing variables (such as process parameters in the parameter set) and the resulting part properties. The values ​​of the main process variables that affect the microstructure and in turn jointly determine the part properties or part quality at the macro level, such as machine configuration, exposure strategy and / or post-processing, can be considered with different weights.

[0194] As mentioned before, this approach is not limited to optimization for a single criterion, but represents the possibility of solving boundary value problems for any thermophysical and manufacturing technology properties. It is possible not only to ensure compliance with the necessary requirements (especially quality requirements), but also to find the most cost-effective production technology method to achieve the given requirements.

[0195] Furthermore, the approach presented here is different from traditional optimization methods (such as those provided by already used topology optimization programs) in that there are multiple options to satisfy local property requirements. For example, a requirement to increase material stiffness locally can be satisfied by adding material, or by adjusting the scanning strategy to produce the desired texture or by replacing the material. From these possibilities, the optimization method presented here always finds a solution on the Pareto Front defined by the boundary value problem.

[0196] Many of the above discussions relate to observations and phenomena applicable to metallic materials, such as inferring properties from crystal texture. Therefore, the method is particularly applicable to metallic materials and is preferred for this purpose. However, in principle, the same or similar methods can be used to establish correlations between selected manufacturing variables and the resulting part properties for ceramics or polymeric materials such as semi-crystalline polymers, so the method can be extended to these material classes with appropriate adjustments. BRIEF DESCRIPTION OF THE DRAWINGS

[0197] The present invention will be described in more detail below with reference to the accompanying drawings by means of embodiments. In this regard, the same reference numerals are provided for the same components in the various drawings. The accompanying drawings show:

[0198] Figure 1 is a schematic partial cross-sectional view of an embodiment of an apparatus for implementing additive manufacturing of the present invention, the apparatus having a control data generating device and a device for generating optimized process variable values, as well as an inspection device and a device for determining property values,

[0199] Figure 2 Schematic diagram of a rod-shaped sample component with two segments and possible layer scanning direction arrangements and their orientations in different layers.

[0200] Figures 3 to 6 For explanation Figure 2Schematic diagram of the arrangement of the layer scanning directions of different layers of the sample component and how their orientation leads to different segment scanning direction distributions of the two segments.

[0201] Figure 7 Schematic diagram for describing another example of a segment scanning direction distribution that is approximately uniformly distributed,

[0202] Figure 8 Schematic diagram for describing another example of a segment scanning direction distribution that is approximately uniformly distributed,

[0203] Fig. 9 is a schematic diagram of an embodiment of an apparatus for generating an optimized process variable value,

[0204] Fig.10 For example Fig.12 Block diagram of possible objective functions for the optimization method shown,

[0205] Fig.11 For the subfunction f S A graph of Fig.12 The possible objective function of the optimization method is to consider the safety factor,

[0206] Fig.12 A flow chart of a possible method flow of an optimization method for generating an exemplary embodiment of a method for optimizing process variable values,

[0207] Fig.13 A perspective view of an example of a component to be manufactured, schematically showing possible forces acting on the component.

[0208] Fig.14 Based on Fig.13 A component with a grayscale representation of the loads acting on each part due to external forces on the component,

[0209] Fig.15 Based on Fig.15 and Fig.14 , showing possible (virtual) segmentation of the component and the Fig.12 The optimization method revolves around the possible definition of the domain of components,

[0210] Fig.16 Based on Fig.12 Flow chart of possible method sequences within method step 3 of the optimization method,

[0211] Fig.17 is a schematic diagram of a first embodiment of a neural network,

[0212] Fig.18 For training purposes Fig.17 A simplified flowchart of a possible approach to a neural network is shown,

[0213] Fig.19 is a schematic diagram of a second embodiment of a neural network,

[0214] Fig. 20 is a schematic diagram of a third embodiment of a neural network,

[0215] Fig.21 is a schematic diagram of a fourth embodiment of a neural network,

[0216] Fig. 22 For training purposes Fig.21 A simplified flowchart of a possible approach to a neural network is shown,

[0217] Fig.23 For training purposes Fig.21 A simplified flowchart of an alternative approach to neural networks is shown.

[0218] Fig.24 is a schematic diagram of a fifth embodiment of a neural network,

[0219] Fig.25 For training purposes Fig.25 A simplified flowchart of a possible approach to the neural network shown, and

[0220] Fig.26 A block diagram of an embodiment of an apparatus for determining a segment attribute value. DETAILED DESCRIPTION

[0221] The following exemplary embodiments will be described with reference to a production device 1 for additively structuring a product in the form of a laser sintering or laser melting device 1, wherein it is once again explicitly pointed out that the invention is not limited to laser sintering or laser melting devices. Therefore, in the following, without limiting the generality, the production device 1 is also referred to as a "laser melting device" 1.

[0222] Figure 1 Such a laser melting device 1 is schematically shown. The device has a processing chamber 3 or a processing area 3 with a chamber wall 4, in which the manufacturing process essentially takes place. In the processing chamber 3, there is a container 5 open to the top with a container wall 6. The upper opening of the container 5 forms the corresponding current working plane 7. The area of ​​the working plane 7 located within the opening of the container 5 can be used to construct the object 2 and is therefore called the construction field 8.

[0223] The container 5 has a substrate 11 which is movable in a vertical direction V and which is arranged on a carrier 10. The substrate 11 closes the container 5 at the bottom and thus forms its bottom. The substrate 11 can be formed integrally with the carrier 10, but can also be a plate formed separately from the carrier 10 and attached to the carrier 10 or simply placed on the carrier. Depending on the specific type of building material, such as the powder used, and the manufacturing process, a building table 12 can be attached to the substrate 11 as a building base on which the object 2 is built. In principle, the object 2 can also be built on the substrate 11 itself, which then forms the building base.

[0224] The basic structure of the object 2 is achieved by first applying a layer of building material 13 to the building table 12, then (as explained later) selectively curing the building material 13 with an energy beam E at the points that form the parts of the object 2 to be manufactured, then lowering the substrate 11 with the help of the carrier 10, thereby lowering the building table 12, and then applying a new layer of building material 13 and selectively curing it, and so on. Figure 1 In the embodiment, the built object 2 in the container on the building table 12 is in an intermediate state below the working surface 7. The built object already has a plurality of solidified layers and is surrounded by the unsolidified building material 13. Various materials can be used as the building material 13, preferably powders, in particular metal powders, plastic powders, ceramic powders, sand, filler or mixed powders or even paste-like materials.

[0225] Incidentally, the working plane 7 is defined here as the x / y plane of the Cartesian reference coordinate system. The z direction runs vertically upward from this x / y plane and forms the main construction direction, since in this direction the base plate 11 descends continuously and the layers L (layers) of the component 2 are superimposed on each other.

[0226] Fresh building material 15 is located in a reservoir 14 of the laser melting device 1. By means of a coating device 16 which is movable in a horizontal direction H, the building material can be applied in the form of a thin layer to the working surface 7 or to the building field 8.

[0227] Optionally, an additional radiant heater 17 can be provided in the process chamber 3. This can be used to heat the applied building material 13 so that the irradiation device for selective curing does not have to input too much energy. This means that, for example, a certain amount of basic energy can be introduced into the building material 13 by means of the radiant heater 17, which is of course still less than the energy required for melting or sintering the building material 13. For example, infrared radiators can be used as radiant heaters 17.

[0228] To achieve selective solidification, the laser melting device 1 has an irradiation device 20, more specifically, an exposure device 20 with a laser 21. The laser 21 generates a laser beam E (as an energy beam E for melting the construction material in the construction field 8). Then, the energy beam E is deflected by a subsequent deflection device 23 (scanner 23) so as to travel along an irradiation path or track in the layer to be solidified according to the exposure strategy and selectively introduce energy. That is, through the scanner 23, the impact surface 22 of the energy beam E is moved on the construction field 8, wherein the current movement vector or movement direction S (or scanning direction S) of the impact surface 22 can be frequently and rapidly changed on the construction field 8. The laser beam E is focused on the work surface 7 in an appropriate manner by a focusing device 24. The irradiation device 20 is preferably located outside the processing chamber 3, and the energy beam E is introduced into the processing chamber 3 via a coupling window 25 installed in the upper side chamber wall 4 of the processing chamber 3.

[0229] For example, the illumination device 20 may comprise not one but a plurality of lasers. Gas or solid-state lasers or any other type of laser such as laser diodes, in particular VCSELs (Vertical Cavity Surface Emitting Lasers) or VECSELs (Vertical External Cavity Surface Emitting Lasers) or an array of such lasers may be preferred here.

[0230] The laser melting device 1 may also include devices (not shown, but known to those skilled in the art) to perform processes such as melt pool monitoring to correct any disturbances that may occur during the manufacturing process in order to maintain as close as possible to the target process control predetermined by the control data created according to the invention.

[0231] The control device 50 here has a control unit 51 which controls the components of the irradiation device 20 via an irradiation control interface 53 , namely sends laser control data LS to the laser 21 , scan control data SD to the deflection device 23 , and focus control data FS to the focus device 24 .

[0232] The control unit 51 also controls the radiant heater 17 with appropriate heating control data HS, the coater 16 with coating control data ST and the movement of the carrier 10 with carrier control data TSD, thereby controlling the layer thickness.

[0233] The control device 50 is connected to a terminal 56 with a display screen or the like via a bus 55 or other data connection. Via the terminal 56, an operator can control the control device 50 and thus the entire laser melting device 1, for example by transmitting process control data PSD.

[0234] In order to optimize the production process, the process control data PSD, in particular the exposure control data BSD in the process control data PSD (both are also synonymously abbreviated as "control data"), are generated or modified in the manner of the present invention by the control data generating device 54, 54', that is, the production device 1 is controlled so that it fully realizes and correspondingly maintains certain optimized process variable values ​​PG0 according to predetermined evaluation criteria during the additive manufacturing process, as mentioned above. To this end, the control data generating device 54 may also include a suitable device 60 for generating optimized process variable values ​​​​PGO, in particular in the form of suitable software or the like. The device as a subunit (such as a software module, routine, object, etc.) may also include a checking device 80 for checking whether a component manufactured using a specific process variable value (expected) meets the property requirements, and a device 70 for determining the property values ​​of such component segments. The preferred procedures for determining the optimized process variable values ​​​​PGO and the preferred embodiments of the suitable device will be referred to later. Figure 2 The following figures are used for explanation.

[0235] For example, the control data generating device 54 can be part of the control device 50 and implemented in the device in the form of a software component. Such a control data generating device 54 integrated in the control device 50 can, for example, receive the required data AD (including geometric data GD) of the component to be manufactured, and on this basis generate the optimized process variable value PGO and appropriate control data PSD based thereon, and transmit them to the control unit 51. The control data PSD especially include exposure control data BSD, but can also include other control data, such as coating control data ST or carrier control data TSD, in order to select a suitable layer thickness.

[0236] However, the control data generating device 54' can also be implemented on an external computer unit, for example, the terminal 56 here, and generate the optimized process variable values ​​PGO and appropriate process control data PSD (in particular, exposure control data BSD) for the part to be manufactured in advance based on the requirement data AD (including the geometric data GD), and then transmit them to the control device 50. In this case, the internal control data generating device 54 in the control device 50 can also be omitted.

[0237] A variant is also possible in which the optimized process variable values ​​PGO are determined in a separate device 60 (e.g. on a separate computer unit connected to the bus 55) based on the required data AD (including the geometric data GD) of the component to be manufactured and then provided to the respective control data generating device, e.g. 54, 54', which only has to determine the appropriate control data PSD, BSD for this purpose. In this way, the control data generating device 54, 54' no longer requires the device 60 for generating the optimized process variable values ​​PGO (or the checking device 80 or the device 70 for determining the component segment property values).

[0238] Several of the above-mentioned options for arranging the various devices 54, 54', 60, 70, 80 in a suitable topology of the computing unit and the control device 50 are described in detail. Figure 1 In addition, other variants can also be realized, for example, the tasks for carrying out the invention are distributed between different computer units or similar devices.

[0239] The process control data PSD, in particular the exposure control data BSD, generated by the control data generating device 54 , 54 ′ can also be regarded as target values, which are then used in the control unit 51 to control the process.

[0240] It should be pointed out again here that the invention is not limited to such a laser melting device 1. The invention can also be applied to any other method for generating or additively constructing a three-dimensional object, in particular by layer-by-layer application and selective solidification of a construction material. Thus, the irradiation device can include not only a laser as described herein, but also any device that can selectively apply energy as wave or particle radiation to or into a construction material. For example, different light sources, electron beams, etc. can be used instead of a laser.

[0241] although Figure 1 Only a single object 2 is shown, but it is possible and usually also common to produce a plurality of objects in parallel in a processing chamber 3 or container 5 .

[0242] For additive construction techniques, as mentioned at the outset, there is a correlation between certain process parameters, such as scanning speed, laser power and scanning strategy, especially in laser melting processes, and the resulting microstructure within the component.

[0243] In crystalline or semi-crystalline solids, for example metal components which are additively constructed using a laser melting process, for example, the crystal texture in particular has a significant influence on the component properties. Texture is defined here as the totality of the crystal orientations. This can be described, for example, by an "orientation density function" (ODF for short). In DE 10 2022 117 935 and DE 10 2022 117 936, microstructures such as texture and the influence of process variables (manufacturing variables) in the component manufacturing process on the texture are described in detail, so these are also mentioned here. However, in polymer or ceramic materials, correlations between selected manufacturing variables and the resulting component properties can also be found, so the invention can in principle also be used for other materials or any construction material.

[0244] The cooling conditions during solidification are key to forming the texture of the component. The key influencing variables here are the temperature gradient that occurs and the feed rate of the solidification front. In laser-based additive construction, the three-dimensional molten pool always exists locally and gradually moves along the scanning direction. Therefore, both the scanning speed and the laser power density have an impact on the texture, because they are also the main influencing factors for the shape and size of the molten pool. For example, at very low scanning speeds, an approximately spherical molten pool is formed, resulting in a heat dissipation direction that is inclined by about 45° to the construction direction. If the scanning speed is increased while the power remains unchanged, the length of the molten pool will increase, while the width and depth (z direction) will decrease, which is why the heat dissipation is approximately aligned along the construction direction (i.e., the z direction) (see, for example, DE 10 2022 117 935). Figure 3 and its description).

[0245] Here, the texture in the component is not only related to the exposure strategy within the corresponding layer, but also to the above-mentioned layer scanning direction arrangement. The layer scanning direction arrangement initially only essentially (jointly) determines the "intra-layer scanning direction distribution" in a single layer. However, since a certain segment of a component or the entire component is composed of multiple layers, the relative position of the intra-layer scanning direction distribution of each layer also plays an important role in the overall texture effect of the segment or component, because different orientations of the layer scanning direction arrangement or the intra-layer scanning direction distribution will also lead to different segment scanning direction distributions, which thus define the frequency of occurrence of each scanning direction in the segment or the component as a whole.

[0246] Figures 2 to 6 An example is given of how two different segments SG2, SG3 of a very simple component 2" generated from multiple layers L produce different segment scanning direction distributions SSV2, SSV3, wherein each segment SG2, SG3 utilizes a different layer scanning direction arrangement HS2, HS3 (filling strategy). The layer scanning direction arrangement HS2, HS3 remains unchanged on all layers of each segment SG2, SG3 and is only rotated by a defined angle between layers (here the angles are different in segments SG2, SG3).

[0247] The component 2" is a simple square rod 2" and the construction direction z extends along the longitudinal direction of the square rod 2", that is, each layer L is oriented along the x / y plane. In the middle area inside the square rod 2", there is an elongated round rod segment SG2. The entire outer area of ​​the square rod 2" except the inner round rod segment SG2 (which forms a kind of core of the square rod 2") is the second segment SG3. Figure 2 Shown on the left.

[0248] exist Figure 2On the right side of the figure, the filling directions in four arbitrarily selected layers L1, L2, L3, L4 (also called layers) of the component 2" are shown to indicate that different layer scanning direction arrangements HS2, HS3 are used in the respective segments SG1, SG2. In the present case, the layer scanning direction arrangements HS2, HS3 each correspond to a very simple filling strategy HS2, HS3, which is used to scan or fill the entire surface of the corresponding segment SG2, SG3. Usually the component is divided into different zones, wherein, for example, the core zone is traveled by a wide track, each track having a specific filling pattern transverse to the track direction, i.e., the filling strategy is much more complicated. In addition, in the edge zone of the component, whether it is the outer edge of the component or the cavity, a contour mode is usually adopted, and the energy beam moves continuously along the contour, so that there is no filling pattern on the surface of the finished component. However, Figure 2 The simplified filling strategies HS2 and HS3 can better illustrate the whole principle.

[0249] As shown by means of the lowest layer L1 (layer shown separately at the side), the inner section SG2 has a filling strategy HS2, in which two tracks always run parallel in one direction, then two adjacent tracks run parallel in the opposite direction, etc. In contrast, the filling strategy HS3 in the outer section SG3 is selected such that one track always runs alternately in the forward direction, the second track runs alternately in the reverse direction, etc. This means that the tracks are running in a serpentine pattern.

[0250] Furthermore, as described above, the reorientation or rotation strategy about the z-axis (main construction direction) of the filling strategy HS2, HS3 is different between the layers for both segments SG2, SG3. For example, in the inner segment SG2, the orientation of the layer scanning direction arrangement HS2, HS3 is always rotated by 45° layer by layer. In contrast, the outer segment SG3 is always rotated by 90°. Figures 3 to 6 As shown, if segments SG2 and SG3 are constructed by a plurality of such superimposed layers, the segment scanning direction distributions SSV2 and SSV3 of the entire segments SG2 and SG3 will be different.

[0251] In these figures, the segment scanning direction distribution SSV3 diagram of the outer segment SG3 is shown above, and the segment scanning direction distribution SSV2 diagram of the inner segment SG2 is shown below. In these and all other segment scanning direction distribution SSV1, SSV2, SSV3, SSV4 diagrams, the frequency of occurrence of the scanning direction in the relevant angle is plotted at angles from 0 to 360°. The reference angle (i.e., the position in the layer where, for example, angle 0° is located) can be selected arbitrarily, since this is only a distribution. For example, the filling direction orientation extending in the x-direction can always be selected as the reference orientation RO of the segment. If the component usually consists of multiple segments, the same reference orientation should be selected for all segments of the component, that is, a reference orientation should be defined for the component. In addition, the frequency of occurrence of the scanning direction can be plotted in arbitrary units.

[0252] Since the scanning paths of the defined layer scanning direction arrangements HS2 and HS3 are relatively accurately maintained here, relatively narrow Gaussian lines always appear in the segment scanning direction distributions SSV2 and SSV3 at the corresponding orientation degrees of the layer scanning direction arrangements HS2 and HS3.

[0253] Between the upper scanning direction distribution SSV3 of the outer segment SG3 and the lower scanning direction distribution SSV2 of the inner segment SG2, each layer is again Figures 3 to 6 The following table shows the Figure 3 The figure shows the bottom layer L1) and the arrow marks show how the various scanning directions of the filling strategy HS3 in the outer segment SG3 of the bottom layer L1 affect the peak in the upper segment scanning direction distribution SSV3, and how the various scanning directions of the filling strategy HS2 in the inner segment SG2 of the bottom layer L1 affect the peak in the lower segment scanning direction distribution SSV2. Therefore, the first layer L1 of the outer segment SG2 will produce a peak at 90° and another peak at 270°. In contrast, the filling strategy HS2 of the inner segment SG2 in the first layer L1 will produce a peak at 0° and another peak at 180°. Others Figure 4 , Figure 5 and Figure 6 It is shown how the layers L2, L3 and L4 located above contribute to the generation of more peaks in the segment scanning direction distribution SSV2 and SSV3 of the outer segment (see the upper curve in each figure) and the inner segment (see the lower curve in each figure). It is clearly shown here that not only the filling strategies HS2, HS3 are responsible for the segment scanning direction distribution SSV, but also, in particular, for the orientation strategy of the respective filling strategy between the layers. The segment scanning direction distribution SSV3 of the outer segment SG3 has peaks only at 0°, 90°, 180°, 270° and 360°, while the segment scanning direction distribution SSV2 of the inner segment SG2 includes significantly more angles.

[0254] However, it is in principle also possible and in practice also preferred to use a significantly more complex or smoother segment scanning direction distribution, in which the scanning directions do not run within such a narrow angle as in the simple embodiment shown in the above figure.

[0255] Figure 7 An example of an almost uniformly distributed segment scan direction distribution SSV3 is shown, where the distribution function is approximated by the probability of being realized in each individual degree direction. Since most machines can typically resolve 1° exactly, the distribution function can be approximated by 360 individual steps.

[0256] This uniform distribution can be achieved in a production structure when a segment consists of many layers and the same layer scan direction arrangement (filling strategy) is used in each layer of the segment, but the orientation of the layer scan direction arrangement from layer to layer is always rotated by an angle that is not a multiple of 360° (for example, the commonly used 67° angle). Then almost all angles appear in the segment scan direction distribution.

[0257] Figure 8 A segment scanning direction distribution SSV4 with an almost uniform distribution angle is also shown. As shown in the figure, such a segment scanning direction distribution SSV4 can also be approximated by basis functions, such as radial basis functions. This has the advantage that the entire segment scanning direction distribution can be parameterized, that is, it can be described by a relatively limited number of free angle distribution parameters, thereby reducing the amount of calculation for finding the optimal segment scanning direction distribution.

[0258] Therefore, it is always possible to change the segment scan direction distribution, for example by selecting other layer scan direction arrangements (i.e. correspondingly changed parameter sets, since the layer scan direction arrangements (unlike the segment scan direction distribution) are also specified as part of the parameter set), in particular other filling strategies, and / or by modifying the orientation or rotation of the layer scan direction arrangements in consecutive overlapping layers, for example by rotating the layers by 45° instead of 90° etc. This has an impact on the texture just as the selection of other process parameters during production has an impact on other properties of the component.

[0259] The present invention can make use of all these relationships, since it can determine or approximate at least one macroscopic property value of the relevant segment based on the known parameter set of the layers used or to be used to construct the component segment and based on the segment scanning direction distribution of the layers for the entire segment. In addition, due to the relationship between the process variable value and the segment scanning direction distribution on the one hand and the expected properties of the product on the other hand, it is possible to determine optimized process parameter values, in particular optimized segment scanning direction distributions in the individual segments and optimal parameter sets (and therefore also optimal scanning direction distributions for the component as a whole) for each segment of the product, so that the component ultimately meets certain (quality) requirement data particularly well.

[0260] Fig. 9 A simplified diagram of a suitable device for generating optimization process parameters is shown. The core of the device 60 is an optimization unit 65 ("optimizer" for short), for example in the form of software. Here, the optimizer 65 contains, for example in the form of a software module, at least one KI-based optimization unit NN, in particular a neural network NN, as a subunit. However, multiple KI-based optimization units NN can also be used in the optimizer 65. Examples of KI-based optimization units in the form of (trained) neural networks and training methods will be explained in detail below.

[0261] The requirement data AD of the intended product can be transmitted, for example, from a user via the requirement interface unit 61 to the optimizer 65. The requirement data AD include at least geometrical data GD of the product, wherein the geometrical data GD can, for example, in the most general case only include the maximum allowed dimensions of the component, or only the maximum or minimum dimensions in certain directions, but on the other hand also include very specific dimensions of certain precise lengths or even CAD data defining the complete contour of the component.

[0262] The optimizer 65 can also obtain hardware property data of the machine used (ie the production device 1 ) via the interface 62 , in particular data on possible process parameters that can be used to control the production device 1 .

[0263] Via the interface 63, the optimizer 65 has access to a property database system DBS (hereinafter also referred to as "database system"), which will be explained in more detail later: in the database system DBS there are certain sets of parameters (in particular scanning speed, laser power density, etc.) which can control the production device 1 during the construction process of the layer, which can be assigned accordingly to the property values ​​of the relevant layer or segment depending on various information about the scanning direction, such as the arrangement of the layer scanning direction within a layer and / or the distribution of the segment scanning direction within a segment consisting of a plurality of layers. For this purpose, this can include, among others (as described above), basic property values ​​BEW for the layers, such as a mathematical description of the texture by the ODF of the layers or their elastic tensor, as well as macroscopic property values ​​MWA, such as a description of the texture or ODF of the entire segment from a macroscopic point of view, and / or macroscopic property values ​​MWA derived therefrom, such as stiffness or strength, to name a few.

[0264] The optimizer 65, for example, uses the following Fig.12 The interpreted program determines the optimized process variable value PGO from all these data and makes it available for further purposes via an interface 64 .

[0265] The entire device 60, i.e. not only the optimizer 65, but also all interfaces 61, 62, 63, 64, can be implemented in the form of software on a suitable computer unit. A database system DBS can also be part of the device 60 and can likewise be implemented on a relevant computer unit. In principle, these interfaces (i.e. the requirement interface 61, the further interfaces 62, 63 and the process variable value interface unit 64) can also be designed as a universal interface unit in order to receive data, process them in the optimizer 65 and output them again.

[0266] The optimized process variable value PGO can be provided by storing it in a suitable memory or sending it to another unit, which then generates optimized control data for the production device based on this value, e.g. Figure 1 One of the control data generating devices 54, 54' schematically shown in FIG.

[0267] The optimizer 65 may also receive information about an expected objective function ZF for the optimization method, wherein the objective function ZF may also be generated at least in part by the requirement data and / or may be obtained from other programs and / or may be specified or configured via a user interface.

[0268] Such an objective function ZF may have a plurality of sub-functions TF1, ..., TFi, ..., TFn (also referred to as "Unterfunktionen (sub-function)" or "Unterfunktionale (sub-function)"), each of which is dedicated to taking into account different requirements. Fig.10 shown.

[0269] Furthermore, with regard to the setting and use of the target function ZF and possible subfunctions thereof, reference is made in particular to DE 10 2022117935. The functions of this document can also be used within the scope of the present invention and are merely supplemented by the present invention.

[0270] A preferred sub-function TF1 may, for example, essentially consist of maximizing the construction rate, and preferably there is also a sub-function TFn which aims at minimizing the variation of the parameter set within the overall structure of the component. This means that the component should contain as few different segments as possible, since each segment is defined in such a way that the layers of the relevant segment are constructed using the same parameter set within the segment. This can be achieved, for example, by means of a sub-function to minimize the number of segment boundaries. There are many other optional TFi sub-functions which may take into account various criteria, such as minimizing the use of material, optimizing the safety indication factor (see equation (8)), minimizing the value of the entropy distribution in the scanning direction of the segment, i.e. reducing the net load or mass of the component as much as possible, the shedding property of the component, etc. and / or any other arbitrary criteria.

[0271] like Fig.10As shown, the objective function ZF is a chain with a (preferably mandatory) first link, which represents the sub-function TF1 for maximizing the construction rate, and with a last link (preferably mandatory in the preferred optimization method with movable segment boundaries to be described later), which represents the sub-function TFn for minimizing the number of segments and thus the parameter set variation (as long as - as preferred - exactly one optimal parameter set is selected for each segment). In between, some optional sub-functions TFi are shown. However, this is only used to illustrate the various possibilities. In fact, the sub-functions TF1, ..., TFi, ..., TFn can be combined into the objective function in any suitable order and manner. In order to determine the priority of the criteria, weight coefficients can also be taken into account in the objective function ZF for the different sub-functions TF1, ..., TFi, ..., TFn. The choice of optional sub-functions depends on the user and his optimization problem and can be expanded as required. By coupling with the boundary value problem or the sequence of mechanical loads or property requirements, the component shape can be optimized in the domain selected by the user for a specific application.

[0272] An objective function F (also called a "quality function" or simply a "function") that can be used as part of an optimization method, by which the best set of parameters for each segment in the previously defined domain Ω and the optimized layer scanning direction arrangement can be determined simultaneously, for example, is mathematically defined as follows:

[0273] F=∫F Seg dΩ (1)

[0274] F Seg is the piecewise objective function of each segment in the domain Ω. The integral is equivalent to the sum of the piecewise objective functions in the domain Ω.

[0275] These piecewise objective functions can be defined as follows:

[0276]

[0277] Therefore, without limiting the generality, the piecewise objective function F Seg Can be described as a subfunction The weighted sum of (Teilfunktionen(sub-functions)), each sub-function is multiplied by the weight coefficient W i , i is the running index for numbering subfunctions, f U The U in the subfunction is just a placeholder for the specific name of the subfunction, for example, U = construction, which refers to the subfunction (Unterfunktional (subfunction))f 构造 Used to minimize construction time or maximize construction rate.

[0278] In principle, all subfunctions f U (Thus the piecewise objective function F Segand the final objective function F) depend in some way on the selected parameter set φ α (x)

[0279] f U (φ α (x)) (3)

[0280] x represents the spatial coordinates in the domain Ω where the optimization is performed (i.e., part and powder segments). This means that a specific set of parameters φ is assigned to each location in the domain Ω. α (x), where the parameter set corresponds to the respective valid parameter set of the segment where the point is located, to construct the layer of the relevant segment. As mentioned above, within the scope of optimization, (in addition to finding the best segment scanning direction distribution) a more appropriate parameter set is selected for a point or segment from a series of candidate parameter sets. α here (and in the following) is an index variable representing various parameter sets φ in the candidate parameter set α (x).

[0281] For example, the subfunction f used to minimize construction time 构造 For example, it can be defined as follows:

[0282] f 构造 =-B α (φ α (x)) (4)

[0283] Subfunction f of the objective function 构造 Can be used to consider each parameter set φ α (x) The impact on construction speed. Sub-function f 构造 The aim is to ensure that for all parameter sets φ that depend on the position x α Among the possible configurations of (x), consider the parameter set with the highest volume construction rate. α Indicates the use of process parameter set φ α (x) The volume construction rate that can be achieved at the respective position x. Subfunction f 构造 Other definitions of can also minimize construction time, as shown subsequently.

[0284] In addition, many subfunctions f U Still depends on the segment scanning direction distribution Ψ(x):

[0285] f U (φ α (x),Ψ(x)) (5)

[0286] The segment scan direction distribution Ψ(x) depends on the position x, since it depends on the segment in which the currently considered position lies.

[0287] A specific example of a subfunction that depends on the segment scanning direction distribution Ψ(x) is a subfunction that aims to adapt the position-dependent stiffness as closely as possible to the stiffness requirements. An example of this can be found in DE 10 2022 117 935.

[0288] As shown in equation (2), users can use a higher weight coefficient W i to emphasize certain requirements in its multi-physics requirement profile, thus ensuring that this aspect is taken more into account when searching for a Pareto optimal solution. In principle, the weighting factor can be any number greater than 0. A wise choice is to always choose a number between 0 and 1, where the sum of the weighting factors can also be normalized to 1. For example, if three sub-functions have to be considered in the objective function, namely one for the safety factor, one for the construction rate and one for the number of segment boundaries, where the safety factor should have a higher importance, the safety factor sub-function would have a weight of 0.5 and the other two sub-functions would have weights of 0.25 each.

[0289] As part of the optimization method, the objective function to be minimized can be defined by combining equations (1) and (2) as follows:

[0290]

[0291] Here the function F has an integral form and is always a scalar value over the entire domain Ω. Therefore, higher values ​​of the quality function F describe less desirable conditions with respect to the requirement profile, while lower values ​​describe more desirable conditions. The optimal solution, i.e. the best parameter set, can be found by minimizing this function (6) The available (candidate) parameter set φ of the distribution Ψ(x) in each optimal segment scanning direction is α (x) OK.

[0292] There are many optimization methods that can be used for this purpose, of which there are two basic cases:

[0293] a) Optimization using fixed segment boundaries.

[0294] b) Optimization with movable segment boundaries, ie the shape of the segment (and thus the component) can be varied.

[0295] In both cases, the optimization is preferably performed in an iterative, sequential manner, wherein all method steps can also be run multiple times in an iterative loop (in particular a nested loop) in order to take into account the influence of the optimization in individual steps on other steps. In principle, it is also possible to perform the optimization with fixed segment boundaries in individual steps and then to run these steps multiple times in an iterative manner, wherein in particular the segment boundaries can also be changed between further steps of the run loop. This means that during each run of the optimization step with fixed segment boundaries, an optimization is performed with possibly changed segment boundaries. This will be discussed later. Fig.12 This specific preferred method is illustrated in more detail.

[0296] However, first the following will outline the basic available optimization methods, either with fixed segment boundaries or with movable segment boundaries:

[0297] a) Optimization using fixed segment boundaries:

[0298] For this purpose, various classical methods, in particular numerical methods, can be used for local or global optimization with and without constraints, linear and nonlinear, where, depending on the form of the objective function F, derivative-free methods are particularly suitable (e.g. the interval halving method, the descent simplex method), methods requiring first-order derivatives (the secant method, the gradient method and the conjugate gradient method, the quasi-Newton method, etc.) or methods requiring second-order derivatives (e.g. the Newton method or the Newton-Raphson method). Depending on the method chosen, the subfunctions must be formulated in such a way that they are valid for the variables to be optimized (i.e. the set of process parameters φ α (x) and / or the segment scanning direction distribution Ψ(x)) are continuous, first continuously differentiable, or even secondarily differentiable. It is preferred to use methods with high convergence, ie those that require as high order derivatives as possible, because such methods are faster.

[0299] Examples of technical implementations of appropriate optimization methods can be found in basic works, such as C. Richter: Optimization in C++: Foundations and Algorithms, Wiley-VCH, Berlin, 2016, where the quality function is denoted by f(x) instead of F and the variable to be optimized is denoted by x.

[0300] In particular in such optimization steps with fixed segment boundaries, it is also convenient to use at least one KI-based optimization unit (as shown below), but the KI-based optimization unit can also be used for optimization with movable segment boundaries in principle. For example, the above-mentioned conventional methods can be used to train the KI-based optimization unit or the neural network.

[0301] b) Optimization using movable segment boundaries:

[0302] Even in the case of optimization with movable segment boundaries, there are various ways to achieve this. As mentioned above, the shape of the segment (and thus the component), i.e. the geometry and the segment scanning direction distribution, can be optimized simultaneously, wherein by moving the segment boundaries, the parameter set applicable to each position will inevitably change, since by moving the boundary, the relevant position may be assigned to another segment, where a different parameter set applies. Without limiting the generality, in principle all methods for topology optimization can be used to minimize the objective function F. These methods include, but are not limited to:

[0303] - Discrete topology optimization, Michell AGM, The limits of economy of material in frame structures. Philosophical Magazine 8(47):589-597, 1904

[0304] -Shape derivative topology optimization, P. Gangl, Sensitivity-based topology and shape optimization with application to electrical machines, Linz, Dissertation 2016 (“Sensitivity-based topology and shape optimization and its application in electric machines”, University of Linz, Dissertation 2016)

[0305] - Level Set, S. Kambampati, C. Jauregui, K. Museth and H.A. Kim, Large-scale levelset topology optimization for elasticity and heat conduction, Structural and Multidisciplinary Optimization volume 61: 9-38, 2020

[0306] - Evolutionary structural optimization, P. Tanskane, The evolutionary structural optimization method: theoretical aspects, Computer Methods in Applied Mechanics and Engineering, 191(47-48): 5485-5498, 2002 (Computer Methods in Applied Mechanics and Engineering, Vol. 191, No. 47-48: pp. 5485-5498, 2002)

[0307] -Phase field, J. Kato, S. Ogawa, T. Ichibangase & T. Takaki, Multi-phase field topology optimization of polycrystalline microstructure for maximizing heat conductivity, Structural and Multidisciplinary Optimization volume 57: 1937-1954, 2018 (Multi-phase field topology optimization of polycrystalline microstructure for maximizing heat conductivity, Structural and Multidisciplinary Optimization volume 57: 1937-1954, 2018)

[0308] For this purpose, the so-called "interface dynamics" must essentially be derived from the target function F, wherein a numerically solvable differential equation is established in which the target function F is derived from the parameters to be optimized. These procedures are generally known to those skilled in the art.

[0309] For the purposes of the present invention, it is particularly advantageous to use the so-called “multiphase field method” as explained in DE 10 2022 117935 (and the proof of principle method therein). However, the present invention is not necessarily restricted to this preferred method.

[0310] The multiphase field method (as a phase field method) is in fact a method for numerically simulating two or more phases and the interfaces between them, i.e. phase boundaries. Phase field methods can be used to determine how the interface structure and shape change over time. Within the scope of the present invention (as described in DE 10 2022 117 935), this principle can be advantageously used to describe the movement of interfaces between adjacent segments, where different sets of process parameters φ should be applied. α (x) and / or the segment scan direction distribution Ψ(x). Therefore, in this example, different process parameter sets φ α (x) and / or the segment scan direction distribution Ψ(x) correspond to different "phases". Otherwise, the procedure can be generally applied.

[0311] To optimize the moving segment boundaries using the multiphase field method, (as described in detail in DE 10 2022 117 935) a nonlinear partial differential equation is usually derived from the objective function F, which determines the movement of the individual segment interface positions (i.e. the orientation of each point or position x on the segment boundary).

[0312] Because on the one hand, there is a parameter set φ at the boundary between two adjacent segments α (x) to another parameter set φ β(x) (β is just another index variable that is not equal to α), but on the other hand, when using the required differential equations, no sharp transitions (sharp interfaces) or jumps are allowed. In the technical implementation of the optimization algorithm, the parameter set φ at position x α (x) Each in its “proportion” The ratio value can be between 0 and 1, where Represents the parameter set φ α (x) exists at position x, and the ratio φ α (x) = 0 means that the parameter set does not exist. Therefore, in the optimization, the position x in the boundary region between two segments (hereinafter also referred to as the "interface region", whose width can be defined by the user) can simply have the first parameter set φ applicable to the first segment α (x) ratio and a second parameter set φ for the adjacent second segment β (x) ratio If more than two segments meet in the interface region, there may be more than two parameter set ratios at position x. In any case, the sum of all parameter set ratios occurring at each position must equal 1. To be able to use parameter set φ in the entire domain Ω α (x) ratio In the optimization, for all positions x outside the middle of a segment, i.e., the boundary of another segment, the parameter set φ α (x) Applicable proportion in this paragraph Simply set it to 1.

[0313] The objective function F or a single subfunction f U The phase field method must be adapted accordingly so that the parameter set φ α (x) ratio This is done separately for the different sub-functions and is described in DE 10 2022 117 935, to which reference is therefore made. This document also relates to the definition and interpretation of the above-mentioned differential equations.

[0314] In the context of the multiphase field method, as explained in detail in DE 10 2022 117 935, it is also possible to use the free angle distribution parameter The segment scanning direction distribution ψ(x) is optimized by optimizing the free angle distribution parameters, wherein the segment scanning direction distribution ψ(x) can be defined accordingly using these free angle distribution parameters.

[0315] For example, in the nonparametric description of the segment scan direction distribution ψ(x), the free angle distribution parameter It can be the ratio of each discrete scanning direction angle in the corresponding segment scanning direction distribution Ψ(x). Therefore, the segment scanning direction distribution Ψ(x) can be decomposed into 360 discrete scanning direction angles, each angle is 1 degree. A free angle distribution parameter Therefore, it is the ratio of the i-th scanning direction angle of Ψ(x) in the segment scanning direction distribution. The values ​​of are between 0 and 1, respectively. The values ​​between 0 and 1 do not represent segment boundaries, but only describe the proportion of the scanning direction angle in the segment scanning direction distribution Ψ(x). At each position x, all segment scanning direction angle components The sum of must equal 1.

[0316] If the segment scanning direction distribution Ψ(x) can be defined parameterically, for example as a Gaussian distribution, then the free angle distribution parameter Alternatively, it may be individual parameters of the segment scanning direction distribution Ψ(x) according to which optimization is to be performed, wherein i and j represent individual parameters respectively (eg, i represents the mean and j represents the standard deviation).

[0317] In practice, existing programs or program fragments for numerically solving such tasks can simply be used to perform optimization using the phase field method. For example, such programs are available in the OpenPhase, OpenFoam or deal.II software packages.

[0318] Since the segment boundaries are defined by the diffusion interface region during optimization with movable segment boundaries, after optimization, it should be determined which process parameter set and which scan direction distribution are finally used in which voxels (Voxeln) in the interface region.

[0319] This depends, among other factors, on the specific use to which the data obtained in the optimization method will be put.

[0320] If this can be used directly to control the production device, different sets of parameters φ known to the voxels can also be used in the interface area α (x) parameter set ratio These parameter sets are respectively assigned to different adjacent segments. In this case, the data of the process parameter sets and their ratios can also be transmitted voxel by voxel to the control device of the production device, for example, during the production process, the process parameter sets are applied multiple times according to their ratios to the overlapping area between two segments. In the laser powder bed fusion process, for example, the laser can be exposed multiple times in the overlapping area, each time using a different process parameter set.

[0321] In order to reconstruct clear segment boundaries in order to visualize the component as a CAD model, appropriate methods such as isosurfaces can be used. Isosurfaces are surfaces connecting adjacent voxels in space that have the same characteristic or a certain value, such as parameter set ratios or free angle distribution parameters. Since, as mentioned above, the segments are preferably also defined by the fact that the same process parameter set φ α This also applies to segments (and, in addition, the same segment scan direction distribution) (in this sense, it can also be called the "process parameter domain"), so the isosurfaces determined in the process should be equivalent to the segment boundaries. α (x) and their respective parameter set ratios In the voxels of , it is necessary to decide which parameter set to apply there. For example, the parameter set with the largest ratio is preferred.

[0322] A method for generating isosurfaces is, for example, the marching cubes method, as described in CD Hansen, CR Johnson, The Visualization Handbook, Elsevier Science, 2005. Other methods from this textbook may also be used.

[0323] It is not absolutely necessary to assign the voxels in the interface region to the segment scan direction distribution accordingly, because the process parameter set is not only defined by the assignment to the segments, but also the specified slice scan direction arrangement is assigned to each segment to which the process parameter set is applied. This automatically results in a segment scan direction distribution for the segment, which should be uniform for the entire segment.

[0324] Some further sub-functions (=Unterfunktionale (sub-function)) are listed below as examples in order to establish the objective function according to equations (1) and (2):

[0325] a) Subfunctions used to minimize the segment interface.

[0326] b) Sub-function to ensure that the part can be de-powdered (applies only to movable segment boundaries).

[0327] c) Sub-functions used to ensure correct heat treatment.

[0328] e) Subfunctions for reducing material usage.

[0329] f) Sub-function for ensuring the best safety factor.

[0330] g) Sub-function for maximizing the change in scan direction angle.

[0331] h) A sub-function used to avoid divergence of the intra-segment scan direction distribution Ψ(x).

[0332] Since the above-mentioned sub-functions are explained in more detail in DE 10 2022 117 935, reference is made here again for the sake of brevity to the explanation of these sub-functions in this document. These sub-functions can in principle also be used within the scope of the present invention.

[0333] However, in the following, reference is made in particular to preferred examples of optimal safety factor assurance with regard to the optimization strategy improvements within the scope of the present invention. Therefore, the subfunctions of the optimal safety factor assurance are further explained (however, these explanations largely correspond to the explanations in DE 10 2022 117 935).

[0334] In practice, the design of structures takes into account the "safety factor" of their loads. The safety factor is expressed as a numerical value and indicates by what factor the failure limit of the material state or the entire component is designed to be higher than the limit that should be determined based on theory. The safety factor is usually determined on the one hand from the material state of the component and the theoretical state variables resulting therefrom, such as strength, and on the other hand from the state of the field quantities acting in the component, such as mechanical stresses.

[0335] In order to reflect this fact in a further development of the method according to the invention, a "safety indicator factor" is preferably introduced, which describes the difference between the set safety factor and the current state of the component or its segment in the simulation. The difference is preferably represented by a number. This reflection can be arbitrary, but preferably at least three states should be represented:

[0336] i) The target safety factor is not achieved,

[0337] ii) fully achieve the target safety factor,

[0338] iii) Exceeding the target safety factor.

[0339] For this purpose, it is preferably defined as follows: a safety indicator value of 0 indicates that the target safety factor S is fully met; a value less than 0 indicates that the safety factor is insufficient; a value greater than 0 indicates that the safety factor is exceeded. In general, the value of the safety factor s is always greater than or equal to 1, otherwise the component is likely to fail under the planned load. The safety factor usually depends on the field of application and, if necessary, its standard. For example, in the automotive industry, typical values ​​for the safety factor s are 1.5 or 2, and in the aviation industry between 1.5 and 6, depending on the safety relevance of the component.

[0340] For example, the parameter set φ for position x α Safety indicator factor Ss of (x) α (φ α (x)) can be defined as follows:

[0341]

[0342] g(σ ij ,φ α (x)) represents the material-specific yield function, which is scaled such that when the mechanical stress σ ij When the yield point of the material is reached, g(σ ij ,φ α (x)) = 1, that is, the component begins to undergo plastic deformation. ij ,φ α (x)) is less than 1, indicating that the component is purely elastic. Therefore, only when the parameter set φ is selected during optimization α (x), so that the material specific yield function g(σ ij ,φ α When the result value of (x)) is lower than the inverse of the safety factor S, the safety indicator factor Ss α (φ α (x)) will be within the “allowed” range and greater than or equal to 0.

[0343] There are various options known to the person skilled in the art for defining an appropriate material-specific yield function. Some variants are published, for example, in J. Betten, Continuum Mechanics, 1993, Springer.

[0344] In principle, suitable material-specific yield functions or their parameters can also be defined with the aid of experiments on suitable samples, such as tensile tests or the like, in particular in the isotropic case.

[0345] According to equation (7), use the "safety indicator factor" Ss α (φ α (x)) of the appropriate subfunction f S can be designed so that for the optimal parameter set φ at position x α (x), safety indicator factor Ss α (φ α (x)) is particularly preferably achieved when the value is equal to 0. Very preferably, it is ensured that exceeding the safety factor is penalized more severely than undershooting, ie the safety factor S is actually met, but the effort required for this is minimized.

[0346] Such a subfunction f for optimization with fixed segment boundaries S The implementation scheme can be as follows:

[0347]

[0348] Here the subfunction is selected in the form of a Leonard-Jones (index 6) potential. α (φ αThe function should show a minimum value when (x) is equal to or close to 0. For values ​​less than 0, the subfunction should quickly take on larger values.

[0349] Using the value of variable A in equation (8), the safety indicator factor Ss can be α (φ α (x)) is shifted on the horizontal axis, where the subfunction f S has a minimum value. The subfunction f in equation (8) S is constructed in such a way that the subfunction f S The minimum value of within the calculation accuracy range for the value A = 0 is Ss α (φ α (x)) = 0.025. Using equation (8) and A = 0, the subfunction f is implemented S This is usually the preferred variant because, in practice, the safety indicator factor Ss α (φ α (x)) can almost never reach the value of 0, but it is safe to ensure that the value is very close to 0, i.e., greater than 0. In a similar way, this can also be achieved by replacing equation (8) with other potential functions.

[0350] For example, in a certain requirement, the safety factor is allowed to decrease within a certain range, but the component volume is required to be as small as possible. In this case, even if the safety factor is slightly lower, the safety indicator factor should be as close to 0 as possible. For example, according to equation (8), the sub-function f S The safety factor S is set to 2, which cannot be achieved for the value A = 0, but should be at least 2.1. However, this case can also be considered with the value A < 0, which on the other hand means that the safety factor in the optimization can also be slightly reduced.

[0351] However, in this case, the safety factor can also be corrected beforehand, for example according to

[0352]

[0353] Execute, thus simply using the modified safety factor S 校正 To replace the safety factor S in equation (7).

[0354] As part of the numerical implementation of the optimization, there may also be subfunctions f S When the value is negative, this is because Ss in equation (8) α (φ α (x))+A becomes negative. In this case, for example, when using equation (8) for optimization, the subfunction f can be simply S Set the value to 10 9, so that the optimization method is forced to choose a different value and thus "correct" the invalid state.

[0355] Fig.11 Show the appropriate subfunction f S , in particular according to the example of the function shown in equation (8). In this subfunction f S The value of (can be any unit; au = any unit) and the safety indicator factor Ss α (can be any unit). It is clear from the figure that from the subfunction f S Starting from the minimum value, with the safety indicator factor Ss α , that is, when the standard is exceeded, the sub-function value increases slowly. However, in the sub-function f S At the minimum value, with the safety indicator factor Ss α Descending (to the left), subfunction f S The value increases dramatically.

[0356] As mentioned above, in order to construct the objective function, preferably at least one minimum configuration is required, which (as described in DE 10 2022 117 935) is particularly preferably composed of a sub-function for minimizing the construction time or maximizing the construction speed and (if the optimization is performed with movable segment boundaries) a sub-function for minimizing the segment boundaries (process parameter interface), i.e. for minimizing the number of segments in the component. In addition, the objective function may also contain other optional sub-functions, such as the other sub-functions mentioned above.

[0357] In the above examples, the simplest form of each subfunction is shown, which can be modified to include more constraints, provided that the constraints in question are not added to the optimization problem in the form of a separate subfunction. Whether the optimization criterion is coupled with another subfunction, especially a forcing subfunction, or whether a separate subfunction is defined depends on the complexity of the optimization problem.

[0358] The following example illustrates the coupling of an optimization criterion with a mandatory subfunction, i.e. the coupling of a safety factor with a subfunction for minimizing the construction time or maximizing the volume construction rate. The subfunction for minimizing the construction time has already been proposed above with the aid of equation (4) (without moving the segment boundaries). In both cases, the subfunction can now be extended with a safety factor to define a subfunction f with a construction rate-safety factor coupling 构造-S :

[0359] f 构造-S =-B α (φ α (x))sign(Ss α ) (4')

[0360] Ssα Here again, the safety indicator factor is represented, for example, as defined by equation (7) above. sign is a sign function that only considers the sign and assigns a positive sign to a value of 0. Therefore, if the parameter set (φ α (x)) leads to a decrease in the safety factor (i.e., the safety indicator factor Ss α is a negative value), due to the subfunction f 构造-S The sign in has changed, and the volume construction rate will no longer be subtracted from the objective function, but will be automatically added to the objective function. This means that falling below the safety factor will inevitably be penalized.

[0361] If you use a sub-function in which the safety factor is already integrated, you do not need to use a separate sub-function to maintain the safety factor.

[0362] The objective function ZF defined in the above manner can now be used in the optimization method (e.g. by Fig. 9 The following will use the optimizer 65 in Fig.12 An example of a possible optimization method is explained. This is an iterative method. The objective function can be used repeatedly in certain method steps, wherein, if necessary, (only) certain sub-functions of the objective function can also be used in different steps, so that the basic optimization objectives of the sub-functions can be initially processed or optimized separately from one another. For example, in a certain step, the effect of certain sub-functions can be reduced or even deactivated by setting certain parameters in these sub-functions accordingly, or certain optimization parameters can initially be considered constant in certain steps.

[0363] exist Fig.12 In the example in , an objective function is used as an example, which contains the following sub-functions, namely: for minimizing the construction time; for minimizing the segment interface; for taking into account the safety factor; for possible part de-powdering; for heat treatment; for maximizing the variation of the scanning angle; for avoiding deviations in the distribution of the segment scanning directions. However, it is once again clearly pointed out that the objective function can also be constructed in other ways, as described above. The best objective function depends on the scope of the requirements, the available computing power and the available time.

[0364] In step S0, a domain G (computational domain or design space) is first defined, which includes the part to be manufactured. If the outer dimensions of the part to be manufactured should not be modified, that is, the shape should remain unchanged, then the outer contour of the part itself can, for example, form the domain. In addition, any frame can be drawn around the part in any way, that is, uncured areas around the part or on certain sides of the part are also counted as part of the domain. The domain is then subsequently (in subsequent steps, see below) divided into several segments, some of which can belong to the part, but some segments (for example powder segments) can also be located outside the part, provided that, as mentioned above, the domain is larger than the part.

[0365] In step S1 , starting values ​​for the subsequent optimization (which runs here iteratively) are then set, namely the starting segment SG′, as well as the starting parameter set PS′ and the starting segment scanning direction distribution SSV′ associated with the starting segment SG′.

[0366] With the help of Fig.13 , 14 and 15 explain how to define a specific component 2' (here, buffer 2'), domain G, and how to determine segments SG0, SG1 in domain G (for example, as starting segments).

[0367] exist Fig.13 In FIG. 1 , the component is shown as a triangular network to show that virtual data can be used to perform finite element simulation on the load condition of the buffer 2 ', where external forces (such as Fig.13 Based on the simulation, a three-dimensional load diagram can be created, which is intuitively displayed in grayscale (or usually color). Fig.14 As can be seen from the figure, for example, only a small volume, less than 3% of the volume of the entire buffer 2' bears load levels above 200 MPa, wherein these higher load areas are mainly located in the crossbeam area of ​​the buffer 2'.

[0368] Knowing the exact load information (and also requirement data, in particular quality requirement data), for example information about areas with greater and lesser loads, the component can be advantageously divided into different virtual segments accordingly.

[0369] Here, the buffer 2' can be divided into individual segments according to the load information, so that the area of ​​the beam with a particularly large load is regarded as a single segment SG1, while the remaining area of ​​the buffer 2' can form another segment. Fig.15 These segments can then be used initially as starting segments SG', for example, in an optimization method.

[0370] Fig.15 It is also shown how, for example, the entire component 2' is surrounded by a larger domain G, and the entire outer region around the component 2' forms a further segment SG0, which is a "powder segment" or "empty segment", in which the powder is not solidified during the building process. For such a powder segment SG0, the starting parameter set can be set so simply in the optimization method that the laser power is 0 here. Then, no further corrections are required for the starting parameter set for the powder segment SG0.

[0371] For all other start segments SG', a suitable start parameter set PS' (for constructing the layer of the relevant start segment SG') and a start segment scanning direction distribution SSV' can then be selected in step S1, for example from a data memory DS, in which various candidate parameter sets KPS can be stored, which can be used to construct the production device 1 to be used. Typically, the number of candidate parameter sets KPS is relatively limited, wherein this number is of course only limited by the available storage space and the computing time for testing the influence of various candidate parameter sets KPS on the property values ​​of the produced components.

[0372] Since high efficiency in component production is also an important criterion in many cases, it is advisable to select each time a starting parameter set PS' and a starting segment scanning direction distribution SSV' which achieve the highest build rate. However, in principle, another selection criterion can also be used. In particular, a suitable KI-based optimization unit NN can also be used to select the starting parameter set. For example, a suitably trained neural network NN can be loaded from a data memory DS for this purpose (if necessary, suitably selected from a database), which will be explained later with the aid of step S3 or more precisely substeps S33 and S34.

[0373] It should be noted here that it is also possible to choose to virtually divide the domain G or the component 2' into the starting segments SG' based on how to achieve the highest construction rate, without using load simulation here, such as Figures 13 to 15 This applies in particular if the component is not subject to high loads at all or if the loads are only a secondary consideration.

[0374] In the subsequent step S2, if the starting configuration defined in step S1, i.e. the starting segment SG', the starting parameter set PS' and the starting segment scanning direction distribution SSV', is used in the manufacturing process, the required simulation of the (still virtual) part to be manufactured is first performed. As described above, for a known configuration or combination of segments SG and the associated parameter set PS and segment scanning direction distribution SSV, macroscopic property values ​​of a single segment at a time can be determined, such as the texture (especially in the form of an orientation density function ODF) and / or other macroscopic property values, such as elastic sensors, yield point distribution, curing coefficient, thermal conductivity, fracture strength, etc.

[0375] In the context of such a required simulation, a load simulation can then be performed, for example using macroscopic property values ​​(of the segments or components formed by the segments), similar to the above-described method with the aid of Fig.14 The simulated buffer 2' is visualized or simulated for vibration or the like. Such simulations can be performed using common numerical simulation methods such as the finite element method or the finite volume method. The result of the simulation is a state description containing different state values ​​of the current system or component and each segment, in particular what loads these segments can withstand, the frequency of the entire system (component), and each time for the current configuration, where the calculation occurs in step S2.

[0376] As will be described later, within the scope of an iterative method, step S2 may be called multiple times to check the current configuration. When the optimization method is first called, these state values ​​or state descriptions apply to the starting configuration starting from step S1.

[0377] In a subsequent step S3, the state description or state value etc. can be compared with external specifications, in particular the requirement data of the component. These external specifications can also include, for example, load records that have been provided in advance for the component as (quality) requirement data, such as Fig.14 Load record of example buffer 2'.

[0378] If, in a special case, all required dimensions are optimally met, then in principle the component can be built using the starting configuration, in particular if this starting configuration has been selected in such a way that the highest possible build rate is achieved. The starting configuration would then be the optimal configuration and the optimized process variable values ​​would have been found. However, the probability of this occurring is very small.

[0379] Typically, if not all requirements are met, the process variable values, ie the segments or their exact segment boundaries, as well as the parameter sets for the individual segments and the segment scanning direction distribution are further optimized in a subsequent method.

[0380] The goal is to assign process parameters to each segment, that is, a complete set of process parameters φ α As well as the segment scanning direction distribution, which offers the greatest optimization potential if, for example, further optimization of the geometry is required, ie, for example, further reduction of mass and / or maximization of the construction rate.

[0381] To this end, in step S3, if necessary, a new current parameter set can be selected from the candidate parameter set KPS of the current segment SG' and, if necessary, a new current segment scanning direction distribution can also be determined. As described above, this selection can be particularly preferably performed taking into account the so-called "parameter set fitness value" PSS (PS score PSS for short).

[0382] Here, different “requirement specific PS scores” can be assigned to the candidate parameter set KPS and / or the segment scanning direction distribution (or a pair of candidate parameter sets KPS and / or segment scanning direction distribution) according to specific requirements, i.e. for each optimization criterion (e.g. strength, stiffness, construction rate, etc.).

[0383] And these parameter set fitness values ​​PSS always depend on the process parameter set φ α .

[0384] Here, the parameter set suitability values ​​with respect to the construction rate and the compliance with the safety factor are shown as an example.

[0385] For the construction rate, the parameter set fitness value PSS can be defined as follows: It is important here that for the proposed embodiment the maximum value of the parameter set fitness value is 1 (=maximum construction rate) and that it is greater than 0 for all suitable parameter set fitness value variants.

[0386]

[0387] Here, the process parameter set B α (φ α ) is normalized by the maximum construction rate of all process parameters available for optimization

[0388]

[0389] The fitness value PSS of other parameter sets depends not only on the process parameter set φ α , also depends on the segment scanning direction distribution Ψ and the current system state in the corresponding segment, such as the homogenized mechanical stress in the above segment An example is the fitness of parameter sets to ensure safety factors.

[0390] The parameter set fitness value that guarantees the safety factor can be as follows:

[0391]

[0392] represents the above safety indicator factor, except that the segment scanning direction distribution Ψ and the homogenized mechanical stress in the segment are shown here ε represents the minimum number that can be represented by the computer. As mentioned above, when the parameter fully meets the current safety conditions, the safety indicator coefficient is 0; when the expected safety factor is not reached, it is a negative number; when the expected safety factor is exceeded, it is a positive number. The exponential function in equation (12) limits the maximum value to 1 as required.

[0393] The requirement specific PS scores PSS can be partially stored in the data memory DS or recalculated for the current configuration. This depends on the specific requirements to which the requirement specific PS scores are related. For requirements that depend only on the selected parameter set, such as the construction rate, these requirement specific PS scores can be stored together with the parameter set. However, for requirements that also depend on external field quantities, in particular mechanical forces, it is preferred to recalculate the PS score each time the loop is run in step S3. An easily understandable example is the mechanical stresses in a component under a given load. These stresses depend, for example, on the geometry of the component and therefore also on the current configuration of the segment. If the boundaries of the segment are changed in the optimization method, the stresses in the component will inevitably also change. Therefore, for such loads, it is best to adjust the PS score each time according to the current configuration.

[0394] First, for this purpose, in step S3, a new parameter set and a new segment scan direction distribution are found for each segment. Hereinafter, the preferred possible procedures of the method flow executed for this purpose in step S3 are described in detail with the aid of Fig.16 the flow chart in

[0395] In the requirement simulation (which can also be referred to as "state simulation") in step 2, the state of each (volume) element (such as a voxel) is determined using numerical methods (such as the finite volume method or the finite element method). Generally, a segment includes a certain number of such elements. Therefore, in the first sub-step S31 of the optimization process in step S3, it is particularly preferable to calculate the homogenized state of the field quantities of the individual elements generated within the segment for each segment. The "homogenized state" is calculated by averaging the field quantities of all elements in each segment in the discrete case. Generally, this operation can be represented by integration and implemented through corresponding discretization in a specific numerical method.

[0396] For example, the homogenized field quantity preferably used in the segment is the homogenized mechanical stress described above (under the specific component requirements or "target conditions" to be achieved, such as a 5 Mpa pressure applied to the side of the buffer). For example, the following integral can be used to determine the homogenized mechanical stress:

[0397]

[0398] This means that: for example, for the homogenized mechanical stress the mechanical stress distribution in the segment is compressed into a value (scalar, vector, matrix or tensor).

[0399] Due to homogenization, the optimization does not need to be carried out for n finite volumes or finite elements, but only for k segments, where reasonably k <= n, and preferably k << n. This can significantly reduce the amount of calculation.

[0400] Here, the homogenized mechanical stress can be represented by a stress tensor as shown in the figure, where for the representation (such as as an input variable of a neural network), it is sufficient to use a six-dimensional stress state vector (the vector contains the matrix elements characterizing the stress tensor in vector form, that is, the diagonal elements σ 11 、σ 22 、σ 33 for compression in the three spatial directions x, y, z, and the elements σ 12 、σ 13 、σ 23 for shear in these directions).

[0401] Another preferred input variable (or input parameter) is, for example, the temperature-time behavior function or a temperature change function (e.g. for a subsequent expected cooling rate and / or a special heat treatment of the manufactured component), where the function can be a simple scalar (e.g. a cooling rate of 300 K / s) or can also be represented by a vector (which contains temperature values ​​at different points in time).

[0402] In the subsequent step S32, a database query is first performed to find a parameter set of particularly suitable "fixed" criteria or requirement parameters, i.e., those requirement parameters that are independent of the segment scanning direction distribution. For example, such a requirement parameter is the construction rate. Thus, for example, a requirement-specific PS score can be calculated for this requirement parameter, and the best candidate parameter sets can then be preselected or ranked based on this PS score for subsequent selection within the scope of further optimization. This can accelerate the search for the overall best parameter set. It is also possible to utilize these steps, for example, within the scope of a KI-based optimization unit or neural network used for training. Then, if necessary, these steps can be omitted in the subsequent optimization (here in step S32), because the KI-based optimization unit has already indirectly taken this into account.

[0403] Subsequently, in step S33, a suitable, i.e. already trained, neural network NN (as a KI-based optimization unit) is searched in a database (e.g. a data storage device DS) for the corresponding field quantities, i.e. those neural networks NN that are designed to be able to use these field quantities as input variables of the neural network NN.

[0404] Here, it is preferred to find neural networks that are able to use different types of required data combinations as input variables and / or generate different types of process variable combinations as output data. An example of such a combined neural network is a neural network that provides a pair of optimal segment scanning direction distributions and an associated optimal parameter set for each segment based on the input stress state. On the input side, the combination can also be as follows, that is, a vector can be input as an input variable into the neural network, which vector includes the stress state on the one hand and the temperature-time change on the other hand (possibly as a scalar in a single vector element).

[0405] The basic structure of neural networks and training methods are already very familiar to the person skilled in the art, so that only a brief schematic overview of possible basic principles of neural networks and possible training methods that can be used within the scope of the present invention is provided below.

[0406] In principle, the neural network NN can be constructed in the form of all known artificial neural network variants. Fig.17 A simple, typical schematic diagram of a first neural network NPS is shown. The neural network NPS is particularly simple, since it has only one input value, namely the homogenized stress state The above six-dimensional vector has only one output value, which is the optimal parameter set PS (given by the scalar φ α , but the scalar represents a fixed tuple of individual parameter values).

[0407] The input vector is usually input to the so-called "input layer" (LI) of the neural network NPS (in the following description, the network NPS also represents other neural networks), which is connected to the "output layer" (LH) through any number of so-called "hidden layers" (LH), and finally outputs the output value in the output layer. Each layer contains a certain number of nodes or neurons, and usually each neuron in the previous layer is connected to all neurons in the next layer, where these connections have different weights. In order to introduce nonlinearity in the neural network (because not all neural network tasks can be represented by linear functions), individual neurons or nodes can deliver results according to the activation function assigned to the corresponding neuron (usually a sigmoid function).

[0408] In a trained network, these weights and the parameters of the activation function ("activation function parameters") are fixed. The number of nodes in the input layer LI depends on the input variable, for example the dimension of the input vector. Similarly, the number of nodes in the output layer LO depends on the output variable. The number of nodes in the intervening hidden layer LH and the number of hidden layers are determined by a person skilled in the art who defines the network for the corresponding purpose before training.

[0409] When training a network, in order to determine the optimal values ​​in the neural network, such as the weights of the node connections and the activation function parameters, for example, training data are used, in which the correct or optimal output variables are known. The "correct" output variables for the input variables can be determined in advance and assigned to the input variables, i.e. using so-called "labeled" training data. Alternatively or additionally, the "correct" output variables for the input variables can also be determined by a solution method running in parallel, for example in "classical" optimization methods. With the help of Fig.18 , Fig. 22 , Fig.23 and Fig.24 Examples of this are provided.

[0410] Therefore, during training, the error can be determined for each output variable, and the weights and activation function parameters can be adjusted layer by layer through error backpropagation and optimization methods to further optimize the network.

[0411] Fig.18 Shown Fig.17 A simplified flowchart of a typical example of such neural network training in . Thus, for the input value (here again for the homogenized stress state On the one hand, we find the best parameter set PS in the classic optimization process NO OR as a comparison value or reference output variable (hereinafter referred to as "reference value"). For this purpose, in the classical optimization process NO, for example, a suitable objective function ZF (for example as described above) and / or a PS score PSS can be used again. Here, in order to make the optimization not only for a certain requirement, but to be able to take all requirements into account, it is preferred to take multiple or all requirement-specific PS scores PSS into account in the method. In particular, for this purpose, the individual PS scores PSS can also be combined into an overall parameter set fitness value PSSG (overall PS score). Fig.18 The optimizer NO is schematically shown to take a request-specific PS score PSS and determine an overall PS score PSSG. The combination of the individual PS scores PSS in the individual segments into an overall PS score PSSG and, if necessary, the sum of the parameter set fitness values ​​over all segments of the component (which can also be used here) will be described later in conjunction with Fig. 22 Detailed explanation.

[0412] On the other hand, the neural network NPS to be trained is used to find the best parameter set as the "prediction value" PSON. The two best parameter sets PS found by different methods OR ,PS ON The comparison is performed in step VG to determine an appropriate error value ERR. For example, the error value ERR may be a mean square error (e.g. based on the representation parameter set PS OR ,PS ON The distance between two vectors in a space determined by the number of vector elements).

[0413] In the subsequent step ES, it is decided whether this error value ERR is small enough. If not, the weights and activation function parameters in the neural network NPS to be trained are corrected by so-called "back propagation", which is Fig.18 In step KB. If the error value ERR is small enough, the method ends in step TE and the neural network NPS is considered to be sufficiently trained. In principle, any other suitable training method can also be applied.

[0414] In the same way, the following neural network NSV can be constructed and trained, which is based on the input value (for example, a six-dimensional vector with stress state ) to find the optimal segment scanning direction distribution SSV. Fig.19 In the example shown in the diagram, the values ​​are ψ1, ψ2, ..., ψ 360 The 360-dimensional vector output is the output value, which represents the segment scanning direction distribution SSV by representing the probability of each angle appearing in a 360° step in the plane, as combined with the above Figure 7As an additional input variable, a preselected parameter set PS can also be used, wherein the input vector can be simply added by adding an additional scalar φ α This trained network NSV can then be used when training the combined neural networks KNSP and KNWS (see, for example, the following discussion of Fig.23 and Fig.25 explanation).

[0415] Likewise, a neural network NNW can be constructed and trained in this way, that is, the neural network is based on other input values ​​(such as the temperature-time behavior function ) to find the optimal parameter set PS (or optimal segment scanning direction distribution SSV). Fig. 20 The method for finding the optimal parameter set PS (given by the scalar φ) is shown. α Schematic diagram of a neural network.

[0416] Fig.21 A first example of a combined neural network KNSP is shown. Here, the best combination of different output parameters is found based on one input variable. Fig.21 In the example of , the input variable is again a six-dimensional vector σ with the stress state ij The neural network KNSP is constructed and trained in the following way: based on this, it searches for the corresponding segment with the segment scanning direction distribution SSV (again with values ​​of ψ1, ψ2, ..., ψ 360 , where the step size and the number of angles can vary, i.e. the vector can have any length) and the associated parameter set PS (again represented by the scalar φ α represents the best combination PSV (i.e. the best pairing).

[0417] This combined neural network KNSP can also be expressed by Fig.18 The method shown is used for construction and training. Fig. 22 To this end, it is sufficient to select or construct a classical optimization process NO' (or "optimizer") in such a way that it can be used as a reference output variable or reference value PSV based on the optimal pairing of the input variable output segment scanning direction distribution SSV and the parameter set PS OR , so as to compare it with the corresponding output variable or predicted value PSV of the neural network to be trained KNSP ON In the manner described above, a comparison is performed in the comparator VG, an error ERR is determined in the process, and the network is then correspondingly further corrected as required.

[0418] To this end, in the classical optimization process NO', in particular, an appropriate objective function ZF (for example as described above) and / or the above-mentioned preferred requirement-specific parameter set fitness value or PS score PSS can also be used again to ensure that the selected segment scanning direction distribution and parameter set best meets the required requirements.

[0419] like Fig.18 As has been schematically shown, it is also possible here to preferably take into account multiple or all requirement-specific PS scores PSS in the method, in particular by combining the individual requirement-specific PS scores PSS into an overall parameter set fitness value PSSG (overall PS score) (in Fig. 22 Schematically again, it is shown that the optimizer NO' uses the PS score PSS and determines the total PS score PSSG).

[0420] When, for example, a single requirement-specific PS score is between 0 and 1, i.e., represents the probability of the satisfaction of a specific requirement with the corresponding candidate parameter set, these requirement-specific PS scores can be simply multiplied to determine the total PS score. For example, if the PS score of the first requirement of the first candidate parameter set is 0.8 and the PS score of the second requirement is 0.2, while the PS scores of the first requirement and the second requirement of another candidate parameter set are 0.6 respectively, then the second candidate parameter set is preferably selected because its total PS score is 0.36, while the first candidate parameter set has only a PS score of 0.16.

[0421] However, this assumes that both requirements are given equal weight. In principle, it is also possible to give a specific requirement a particular weight. This can be taken into account using a weighting factor when determining the overall score for the PS.

[0422] When only considering the PS score used to ensure the safety factor and PS score for construction rate In the case of , for example, the following product can be obtained:

[0423]

[0424] By maximizing the total PS score One can try to determine for each possible parameter set a segment scan direction distribution in the corresponding segment, which can be used in pairs to maximally satisfy all criteria:

[0425]

[0426] That is, by changing the segment scan direction distribution, the total PS score is calculated with a fixed selected (candidate) parameter set. The maximization of φ is performed for all possible (candidate) parameter sets of the segment. Then, the best parameter set φ can be selected for the corresponding segment from all (candidate) parameter sets with corresponding best segment scanning direction distributions. Opt :

[0427]

[0428] Therefore, for nested optimization, the following optimization formula is obtained:

[0429]

[0430] So far, the above schemes are limited to one segment.

[0431] To find the best output variable for the entire component (i.e., the appropriate pairings of parameter sets for individual segments and segment scan direction distributions that ultimately achieve the best component as a whole), for example, all total PS scores over all segments can be formed The sum of (i.e. the sum of the fitness values ​​of the parameter sets). For this purpose, it is suitable to use a weighted sum, for example, which needs to be maximized:

[0432]

[0433] V s This indicates the volume of the corresponding segment.

[0434] For this optimization method, a heuristic approximation method can be preferably selected, particularly preferably a simulated annealing method (SA method) or a quantum annealing method (QA method). For simplicity, the optimization by means of a heuristic approximation method, in particular an SA method or a QA method, can be performed separately in each segment, since for the sum, the commutative law applies and the sum of the partial maxima must result in a maximum value. These methods are superior to the classical optimization methods NO, NO' for training neural networks (e.g. Fig.18 and Fig. 22 It is also very convenient.

[0435] Combinatorial problems are solved using SA or QA methods. They are used to find approximate solutions to optimization problems that, due to their high complexity, exclude the exhaustive attempt of all possibilities and mathematical optimization methods. The goal is to find the optimal solution from which a coupled geometry and scanning strategy optimization with the greatest optimization potential can be carried out. These methods take their name from the mathematical simulation of cooling processes, such as annealing in metallurgy. After heating a metal, the atoms have enough time to align themselves during the slow cooling process and form a stable crystal. In this way, a state with the lowest possible energy (close to the optimum) is achieved. For the SA or QA methods, the temperature corresponds to the probability that an intermediate result of the optimization may become worse. Unlike the local search algorithm, this method can deviate from the local optimum again. Poor intermediate solutions are accepted because this provides the opportunity to find a better local optimum, i.e. in this case a result with a better overall PS score.

[0436] However, both methods are known in principle (see, for example, the solution to the typical "traveling salesman problem") and therefore do not need to be described in detail here. Suitable methods are described, for example, in "An Effective Simulated Annealing Algorithm for Solving the Traveling Salesman Problem" by Wang, Zicheng et al., Journal of Computational and Theoretical Nanoscience, Vol. 6, No. 7, pp. 1680-1686 (7), July 2009, for the classical variant, and in "Quantum annealing of the traveling-salesman problem" by Roman Martoňák et al., Phys. Rev. E, Vol. 70, 10 November 2004, 057701, for the QA method.

[0437] In order to further speed up the method, a certain preselection of suitable candidate parameter sets can first be performed again using a KI-based method. This reduces the space of selectable parameter sets, each of which must determine an optimized segment scanning direction distribution. In practice, the field quantities are inherently limited. For example, if the permissible mechanical stresses are exceeded, the component will never exist. Under purely elastic loads, this permissible range of values ​​can be described by a fluid. In this case, in a previous training, this permissible space can be evaluated at discrete points and these points can be optimized. If such a procedure is used in other classical optimization methods N0, NO', then a hybrid method is actually used to create reference values ​​in order to then train a more complex neural network.

[0438] Finally, for each segment of the part, there is a pair of optimal parameter set and optimal segment scanning direction distribution as the reference value PSV OR , used to compare with the predicted value PSV ON For comparison, this prediction value is sought or has been found by the neural network KNPS to be trained.

[0439] Alternatively, this optimization method can be performed in advance and the optimal scanning angle distribution (optimal segment scanning direction distribution) is calculated for each segment for all possible (candidate) parameter sets and stored in a lookup table. The values ​​in the table can then be used as labeled training data to train the neural network.

[0440] In (especially with Fig.18 and Fig. 22 In the above-mentioned training methods (in which relatively classical and therefore relatively complex methods are used to form training values ​​or reference values), it is necessary to take into account that the training can also be carried out in such a way that the individual neural networks to be trained are first "pre-trained" with less effort using classical methods (e.g. the methods work only in segments, i.e. the sum of the fitness values ​​of the parameter sets is not used); and then further training (i.e. a kind of "fine tuning" or so-called "transfer learning") is carried out, using more complex methods to create reference values, for example using the sum of the fitness values ​​of the parameter sets, in order to take into account the optimization of the entire component. If different components with similar (standard) segments are constantly being created, it is also possible to consider storing the pre-trained neural networks of these segments in a database and then individually "retraining" the corresponding components using more complex optimization methods.

[0441] It is also possible to pre-train a KI-based optimization unit or a neural network for a specific set of requirements and / or process parameters and then retrain it individually for the respective current requirements or process parameters by transfer learning. For example, a neural network trained considering a specific steel type as a construction material can be quickly retrained to adapt to other similar material types.

[0442] If the combined neural network KNSP trained using the above optimization method or the guidance data generated thereby is used in a subsequent optimization method (e.g. Fig.12 If used in step S3 of the method shown, the optimized process variable values ​​of the component are finally determined so as to determine optimized process variable values ​​for each segment, which are optimized within the corresponding segment for the total parameter set fitness value and in the component for the sum of the parameter set fitness values.

[0443] Fig.23 A flow chart of a possible method is shown in order to be able to train such a combined neural network KNSP as quickly as possible, which is based on the stress state As input variables for the respective segment the best combination of segment scanning direction distribution SSV and parameter set PS is again provided as output variable.

[0444] Similar to Fig. 22 The program shown here will also enter the variable Provided to the neural network NPS to be trained, the neural network itself determines the best combination of the segment scanning direction distribution SSV and the parameter set PS. This pairing is used as the prediction value PSV' ON For further comparison with reference values. However, in order to determine the reference value PSV' OR , do not use Fig. 22 Instead of using the complex, special, classical optimization method shown in , we use two simple neural networks that have already been trained (one based on the input value Determine the optimal parameter set PS ON The network NPS is based on the input value and based on the best parameter set PS found by the first network NPS ON Find the trained neural network NSV of the relevant optimal segment scanning direction distribution SSV), so as to finally determine the optimal combination of parameter set PS and segment scanning direction distribution SSV as reference value PSV' OR Here, the two best parameter sets PSV' found by different methods OR , PSV' ON This comparison will also be done in step VG to determine an appropriate error value ERR (e.g. the least mean square as described above). In a subsequent step, it will again be decided whether this error value ERR is small enough. If this is not the case, the weights and activation function parameters in the neural network are corrected again by so-called "back propagation", which is done in Fig.23 This is again indicated by step KB. If the error value ERR is sufficiently small, the method ends in step TE and the neural network is considered to be sufficiently trained.

[0445] Of course, the premise is that there are already trained neural networks NPS, NSV for single values. Here, it should be noted that in order to determine the optimal segment scanning direction distribution, the network NSV must be constructed and optimized in the following way, that is, based on the stress state and the preset optimization parameter set as input variables to find the optimal segment scanning direction distribution. In other words, the network is actually a combined neural network here, but two different types of input variables are used here to find the output value (segment scanning direction distribution).

[0446] It should also be noted that: here, when creating a reference variable or reference value PSV' OR When the optimal parameter set PS pre-selected in the first neural network NPS is input ON , and the second neural network NSV uses this fixed value and only finds a suitable segment scanning direction distribution SSV for it. In other words, different from Fig. 22 The method shown in Fig.23 In the method of , a number of possible combinations are no longer tried in the training phase, but the segment scanning direction distribution SSV is first optimized for all possible parameter sets and then it is checked which pairings form the best overall PS score overall. Accordingly, this method is significantly faster when training the neural network KNSP, and the result of this combined neural network KNSP is completely sufficient for the entire method, because the output variables found (i.e. the pairings of parameter sets and segment scanning direction distributions) are only first approximate solutions and are usually used in subsequent method stages (see Fig.12 ). However, the subsequent method steps of the entire method are improved by finding a better starting point. In particular, the method converges faster, i.e. the subsequent optimal solution is found faster.

[0447] Fig.24 A schematic diagram showing an example of a more complex combined neural network KNWS, and Fig.25 For this purpose, a flow chart of a possible training method is shown.

[0448] According to Fig.21 The neural network KNWS is constructed and trained in the following way, that is, it outputs the segment scanning direction distribution SSV (with values ​​of ψ1, ψ2, ..., ψ 360 ) and the associated parameter set PS (again represented by the scalar φ α The best combination of (expressed) is used as the output variable PSV". However, Fig.21 The difference between the neural network KNSP and the stress state KNSP is that in this case, the input variable is still a combination of two different types of required data. Therefore, a seven-dimensional vector is used as the input variable at the input layer LI. The seven-dimensional vector includes the stress state and contains additional scalar values ​​as further vector elements, which represent the function of the temperature-time behavior (For example, here it is a simple cooling rate, the unit is K / s).

[0449] Fig.25 The simplified flowchart shown shows a possible method to train such a combined neural network KNWS as quickly as possible, which is constructed in the same way as Fig.23 The method shown is very similar. Here, the key point is still to use the already trained simple neural network NNW, NPS, NSV to determine the best combination of parameter set PS and segment scanning direction distribution SSV as reference value PSV" OR .

[0450] Therefore, on the one hand, a trained neural network NPS is used, which is based on the input values The stress-optimized parameter set PSσ is determined and on the other hand a trained neural network NSV is used, which is based on the input values and the pre-optimized parameter set PS found by the neural network NPS OR To determine the best segment scanning direction distribution SSV. In principle, the neural network NPS and NSV can be based on Fig.23 The same network used in the method.

[0451] In addition, however, another trained neural network NNW is used here, based on the temperature change function As input variables to determine the parameter set for temperature optimization

[0452] Thus, the stress-optimized parameter set PSσ and the temperature-optimized parameter set First, it is provided as an input variable to the parameter set selector PAS. In the parameter set selector, it is decided which parameter set is used as the parameter set input variable PS. ON is transmitted to the neural network NSV to determine the optimal segment scanning direction distribution SSV. For this selection, the parameter set selector PAS can preferably use the so-called "strategy reinforcement learning method". Then, the neural network NSV can output the optimal parameter set PS ON The pairing with the best segment scanning direction distribution SSV is used as the reference value PSV" ORSuitable “policy reinforcement learning methods” are known to the person skilled in the art and can be found, for example, in the following literature, namely: “Learning to Optimize” by Ke Li and Jitendra Malik (arXiv:1606.01885, 2016 and International Conference on Learning Representations (ICLR), 2017); or “Learning to Optimize (Learning to Optimize Neural Networks)” by Ke Li and Jitendra Malik (arXiv:1703.00441, 2017).

[0453] At the same time, here, enter the variable is also fed again to the neural network KNWS to be trained, which itself provides the best combination of the segment scanning direction distribution SSV and the parameter set PS as the prediction value PSV" ON , for step VG with reference value PSV" OR Make further comparisons.

[0454] Based on the error value ERR determined here (e.g. the above-mentioned least mean square), it can then be decided again (in block ES): whether this error value ERR is small enough and whether the neural network KNWS is considered to be sufficiently trained (block TE), or whether it is necessary to further correct the weights and activation function parameters of the neural network KNWS in block KB.

[0455] In addition, according to Fig.25 The method has the following fact: here, when creating a reference output variable or reference value PSV" OR When inputting a fixed pre-selected optimal parameter set PS ON , and the second neural network NSV uses this fixed value and only searches for a suitable segment scanning direction distribution SSV for it. As mentioned above, the output variable PSV" subsequently found by the combined neural network trained in this way is only a preliminary approximate solution and is usually further modified in subsequent method steps (in most cases).

[0456] If a suitable neural network NN is found, it can be used in ( Fig.16 In the method shown in the figure, in step S34, the data of the neural network NN is loaded into the database, specifically, the weights of the node connections between the different layers LI, LH, LO and the activation function parameters of each node of the trained neural network are loaded. It should be mentioned here that the reference symbol NN in the figure can represent any KI-based optimization unit NN suitable for a specific goal, and in particular also represents the above-mentioned neural networks NPS, NSV, NNW, KNSP, KNWS.

[0457] Using the data of the selected neural network NN or the selected neural network NN, a corresponding optimal segment scanning direction distribution and an associated optimal parameter set can subsequently be quickly calculated in step S35 based on the input data.

[0458] Then, in the next step S36, the current segment scanning direction distribution and parameter set in the relevant segment are updated in the method flow accordingly for further optimization, and then according to Fig.12 The method works using these new segment scan direction distributions and parameter sets.

[0459] Then, at the end of step S3, identical segments SG' may still exist, but some segments SG' should preferably be assigned a more optimal current parameter set to better meet the requirements.

[0460] After step S3, in step S4 (see further Fig.12 ) uses the objective function ZF to optimize the segment boundaries, i.e., attempts to obtain better results by moving individual segment boundaries of certain zones. If necessary, a KI-based optimization unit or a neural network NN can also be used to assist this operation. This also explicitly includes that not only the segment boundaries of the segments within the component are moved, but also the segment boundaries between the segments at the edge of the component and the powder segments outside the domain may be moved. This means that in some cases, the outer contour of the component can also be changed, such as some pillars becoming thicker or thinner, depending on the needs of the specific situation. In this way, the geometry of the component can be optimized at the same time.

[0461] Optimization possibilities in step S4 are also explained in DE 10 2022 117 935, wherein not all possibilities described in this document necessarily have to be used.

[0462] At the end of step S4, the improved segments (and optionally further improved segment scan direction distributions and parameter sets) may match the geometry and segment boundaries of the parameter set respectively selected in step S3.

[0463] In step S5, the process in step S2 is repeated again, that is, a new state description (also synonymously called system description) is created using the current process variable values, i.e. the current segment, the current parameter set and the current segment scanning direction distribution, and it is determined and checked whether all requirements, especially quality requirements, are fully met.

[0464] If the requirements cannot be fully met, the system will jump back to step S4. The loop runs between steps S4 and S5 until the termination criterion is reached, that is, for example, until the change between two iteration steps with respect to the predetermined quality criterion becomes very small. At this point, it can be considered that there is a nearly optimal combination available for the current load situation.

[0465] In a subsequent step S6, which includes three sub-steps S6a, S6b and S6c, it is checked whether all the zones where powder is present also have an outlet from the component. Thus, at least for those cavities in the component that are not intentionally required to be filled with powder, it should be ensured that powder does not remain in the component after unpacking, for example in cavities that are not connected to the outside space.

[0466] For this purpose, in step S6a, it may be assumed that the powder is a viscous fluid flowing out of the cavity.

[0467] By jumping back to step S4, in which the objective function ZF is also used to correct the segment boundaries, the segment boundaries can then be changed so that the areas with powder inclusions can be minimized or completely eliminated. This can be done in a loop, which will try for a certain number of iterations to move the inclusions so that they end up on the component surface by changing the segment geometry or by filling the inclusions with molten material, i.e. eliminating the cavities filled with powder. For example, a termination criterion can be used here again, i.e. no more relevant changes are made in the loop or a maximum number of iterative steps have been completed.

[0468] Subsequently, in an optional step S6 b , a so-called Minkowski subtraction can be performed on those regions in which powder inclusions may still be present, in order to remove these regions from the calculation grid by erosion similar to image processing methods.

[0469] In the final step S6c, it is checked whether there are still powder inclusions. If so, it jumps back to step S3 to clear these areas. There, a new parameter set is selected for the relevant area, so that this area is consolidated, and then the complete optimization is carried out again starting from step S3 with the new parameter set.

[0470] However, steps S6a to S6c are explained in more detail in DE 10 2022 117 935, since reference is also made thereto in this regard.

[0471] It should be noted that the de-powdering step S6 is intentionally performed separately after the optimization of other points in the objective function in step S4. This can be achieved by setting the pressure everywhere to 0 in the first run, and therefore all other criteria are first optimized in the previous run of steps S4 and S5 without de-powdering. Therefore, with respect to the de-powdering criterion, the objective function ZF or the corresponding sub-function is initially set to 0 at all locations in the domain due to a clever choice of parameters. This procedure can save computational time if, when starting the optimization in the starting configuration, the shape of the initially available solution is far from the optimal shape, so that a large number of iterative cycles between steps S4 and S5 are expected.

[0472] Steps S7 and S8 are purely optional and can be used to compensate for errors that may have occurred in the previous steps due to the homogenization of the segment states (e.g., determining the average state of each segment even if the state of each segment varies spatially), since this homogenization may contain certain errors. However, by combining the homogenization with the KI-based optimization, at least a very high-quality approximation of the optimal solution close to the actual optimal solution can be achieved anyway, so that usually only a small amount of "subsequent optimization" (similar to "fine-tuning") is required. In this process, step S8 corresponds to step S5 or S2, that is, a state description and check is performed here to determine to what extent the system or component can meet the requirements of the current segment and the parameter set currently assigned to the segment. If the requirements cannot be fully met, a jump back to step S7 is made. The loop between steps S7 and S8 will be performed again until the termination criterion is reached, that is, for example, until the change between two iterative steps with respect to the predetermined quality criterion becomes very small. For the specific process of steps S7 and S8, reference can also be made to DE 10 2022 117 935. In this case, in steps S7 and S8, possible spatial variations of the state (e.g. mechanical stress) within the segment are also taken into account, so these steps are of course more complex than steps S2 and S5. However, since it can be determined that the current state is very close to optimal when starting these steps S7 and S8, only a small number of iterations are required here.

[0473] Step S9, which is also optional, ultimately concerns the possible heat treatment of the component to be manufactured later (if the heat treatment has not been fully taken into account in step S3 with the help of the neural network). This step includes two sub-steps S9a and S9b. In sub-step S9a, the virtual component to be manufactured (still) is subjected to a virtual heat treatment and the characteristic temperature curve of each point in the simulated heat treatment is stored. The subsequent sub-step S9b will check whether the simulated temperature curve is within the allowable range of the required heat treatment, for example whether it becomes too hot or not hot enough at certain points of the component. If the limit value is exceeded, it returns to step S2 and finally the entire optimization is carried out again with a new starting configuration, where the starting configuration should be selected so that the heat treatment problem may be eliminated. On the other hand, if the requirements for heat treatment are finally met, the optimization process is finally terminated and the required optimization process variable values ​​PGO in the following form can be obtained, namely the optimal segment boundary SGG, the optimal parameter set PS and the optimized segment scanning direction distribution SSV.

[0474] The optimization of the segment boundaries SGG can also include an optimized alignment of the object relative to the main construction direction, i.e. the Z direction of the stacking between the layers. The segment boundaries can also be corrected to achieve a reorientation or optimized orientation of the components relative to the main construction direction. For example, a suitable orientation in the installation space can reduce or minimize overhangs and / or supports. Reference is also made to DE 10 2022 117935 for further explanations in this regard.

[0475] Finally, it should be noted that within the scope of the optimization process, all segments of the component are preferably optimized simultaneously, that is, for example, for all starting segments SG', at the beginning of step S1, not only the starting segment boundaries are determined, but also other starting parameter values ​​PS' and the starting segment scanning direction distribution SSV' are determined and always optimized together in the corresponding step. This is particularly convenient in the case of the invention using a KI-based optimization unit, because in this way the optimization can be performed significantly faster than with conventional optimization.

[0476] exist Fig.12 In the optimization method shown, as described above, each current configuration is evaluated, for example, in steps S2, S5 and S8. Here, it is checked whether a construction process using the segments currently present in the optimization method (i.e. the current segment boundaries) and the current parameter sets belonging to the segments and the current segment scanning direction distribution SSV will result in a component that meets certain requirements. This means that a state description of the virtual component can be determined using a state simulation and, if necessary, compared with predetermined (quality) requirements in a further step.

[0477] The macroscopic property values ​​of each segment can be used to determine the state or determine the state description. As mentioned above, these macroscopic property values ​​can be in particular the texture in the segment, which can be described by the above-mentioned orientation density function ODF, as mentioned above, and other macroscopic property values ​​derived therefrom, such as elastic sensors, yield point distribution, hardening coefficient, thermal conductivity, fracture strength, etc.

[0478] Now with the help of Fig.26 It is explained how, using a known set of parameters PS for constructing the layers of a segment and a known segment scanning direction distribution SSV of the segment, a macroscopic property value MWA for the relevant segment is determined in each suitable device 70 or unit for determining macroscopic properties.

[0479] It should be clearly pointed out that the device 70 can also be implemented in the form of software on a suitable computer unit. In particular, it can be integrated into the optimization method, for example as a software object or subroutine. Similarly, all other components of the device 70 described now, such as interfaces and database systems, can also be implemented in software. In addition, the interface can also be implemented partially by hardware and partially by software. For example, the entire device 70 can also be distributed on different computer units, which are connected to each other in a suitable manner. This applies in particular to the database system DBS used by the device 70, which includes, for example, a macro attribute database EDA and a basic attribute database EDB, which can be easily outsourced to other computers and storage units. The functions and data contents of the macro attribute database EDA and the basic attribute database EDB, as well as the options for setting up such databases EDA and EDB will be explained below.

[0480] For example, the current parameter set PS can be adopted via the parameter set interface unit 72, and the current segment scanning direction distribution SSV can be adopted for the segment construction process via the scanning direction interface unit 73. In addition, the device 70 can have an interface 74, through which segment information SGI, i.e. information about the segment, such as the number of layers, the current segment boundary, etc., can be adopted.

[0481] All this information can be used in the macro attribute determination unit 71 to determine the macro attribute value MWA, or more preferably to determine the whole set of macro attribute values ​​for the segment concerned, and to assign the current parameter set PS and the current segment scanning direction distribution SSV as well as the segment information SGI to these macro attribute values. The macro attribute determination unit 71 works in the following manner: Fig.26 The macro-attribute determination unit 71 is shown in a greatly simplified form in the form of a flow chart.

[0482] In a first step MS1, the macro attribute database EDA may first be queried to determine whether a ready-made macro attribute value MWA has been stored for a specific combination of parameter set PS and segment scanning direction distribution SSV. If so, the macro attribute value MWA is simply adopted and may be returned from the macro attribute determination unit 71 via the interface 75 of the device 70 to, for example, a higher-level software component, which then continues to use the macro attribute value MWA.

[0483] Preferably, particularly frequently occurring macro-attribute values ​​MWA of combinations of parameter sets PS and segment scanning direction distributions SSV are stored in the macro-attribute database EDA, ie these are standard combinations that are repeatedly used. Of course, the macro-attribute database EDA can be expanded step by step.

[0484] If the query in the macro attribute database EDA is unsuccessful, a new macro attribute value MWA must be determined for the current individual case based on the current parameter set PS and the current segment scanning direction distribution SSV. To this end, in the next step MS2, the current basic attribute value BEW of each layer of the current parameter set PS is first queried in the basic attribute database EDB. For example, the basic attribute value BEW can be the texture and / or microstructure MS of the layer, or a value derived therefrom that is applicable to each layer. However, it is preferred to continue to use the texture TX and microstructure MS described by the ODF.

[0485] In the third step MS3, the basic attribute values ​​BEW of each layer are mathematically homogenized, that is, the basic attribute values ​​BEW of each layer in the segment are appropriately combined to approximate the macroscopic attribute value MWA of the complete segment. The information used here includes the number of layers, the arrangement of the layer scanning directions in the layer, and the relative rotation between the layers, which determines the current segment scanning direction distribution.

[0486] For example, within the scope of the homogenization process in step MS3, it is possible to simply form an average value of the basic property values ​​of the individual layers, and then form the desired macroscopic property value MWA from this average value. Alternatively, the reciprocal value of the average value of the basic property values ​​BEW of the individual layers can be determined first, and then the reciprocal value of the average value of these reciprocal values ​​can be formed. The reciprocal value of this average value forms the macroscopic property value. Which of these two methods is used depends on the appearance of the microstructure MS of the individual layers and the current load requirements.

[0487] It should be pointed out here that, as mentioned above, the basic property values ​​BEW of the individual layers do not differ significantly as long as they are manufactured with the same parameter set PS (i.e. the same filling strategy), except for the fact that the orientation of the basic property values ​​changes with the orientation of the layers relative to the (in principle arbitrarily defined) reference orientation RO. This naturally leads to a change in the orientation of the texture TX. Ultimately, this also has an impact on all property values ​​in the form of orientation-dependent material parameters, such as the elasticity tensor or the yield point distribution, for example in the form of the Hill tensor, which can differ significantly in different orientations. However, it is sufficient to know the basic property values ​​for one orientation, preferably the reference orientation. The basic property values ​​for other orientations can be calculated using simple operators, such as simple rotations.

[0488] Then, the macroscopic property value MWA determined in step MS3 can be output again via the interface 75, for example, to a higher-level unit, and then further processed by the higher-level unit.

[0489] In addition, the macro-attribute value MWA can also be stored in the macro-attribute database EDA together with the parameter set PS on which the calculation is based and the associated segment scanning direction distribution SSV. If the macro-attribute database EDA has enough space, in principle each new macro-attribute value MWA can be stored in the macro-attribute database EDA. However, for example, very rare parameter sets PS or segment scanning direction distribution SSV, etc., it is preferably not necessarily to do so. In principle, the system can also be designed to be learning-type, that is, for example, it is noted in a list which parameter combinations PS and SSV appear particularly frequently, and then the macro-attribute database EDA is gradually expanded for these parameter combinations, or vice versa, each macro-attribute value MWA is first stored in the macro-attribute database EDA, and if it is no longer queried for a period of time, it is deleted again to create storage space for other combinations.

[0490] DE 10 2022 117 935 and in particular DE 10 2022 117 936 describe in detail the structure of the basic property database EDB with the aid of a number of figures, for example: different test specimens are produced and measured in different test production methods to determine basic property values ​​BEW of one or more layers of the test specimens obtained with different parameter sets, so that reference may be made to these documents. The contents of these documents are therefore also incorporated herein.

[0491] As described above, the method and apparatus for determining a segment attribute value or checking whether a segment current state satisfies a certain condition can be used in particular in an optimization method to determine appropriate process variable values ​​for product production.

[0492] However, in principle, such a check can also be performed completely independently of the optimization method, for example, by checking control parameters before use, which are used to manufacture a component, but in a different manner than the optimization method described above. It is also possible to perform a subsequent check on already manufactured components, which should not be destroyed and therefore cannot be subjected to a specific load test. For this purpose, it is sufficient to know the process parameters used in production and required for the method described above.

[0493] For example, a possible structure of an inspection device 80 that can be used for this purpose as well as an inspection method are described in detail in patent application DE 10 2022 117 935 (see also Fig.24 and Fig.25 and related instructions), the contents of which may be referred to herein or deemed to be incorporated herein. The inspection device and inspection method described in the document are also applicable to the device and method described in the present application.

[0494] Finally, it should be pointed out again that the apparatus and method described in detail above are only embodiments, and various modifications may be made to them by those skilled in the art without departing from the scope of the present invention. In particular, the optimization method can be adjusted almost at will according to current requirements, for example, additional steps can be added, or steps can be merged, or the optimization criteria can be exchanged or expanded. The optimization criteria can also be considered in different ways. It should also be noted at this time that although the above-mentioned method of using the weights and sums of sub-functions to form the objective function is preferred, this method is not necessarily limited to this. For example, the sub-functions can also be defined in the form of constraints, such as using the Lagrange multiplier method. The constraints can be, for example, equality or inequality constraints. An explanation of this can be found in basic works such as C. Richter "Optimierung in C++: Grundlagen und Algorithmen (Optimization in C++: Foundations and Algorithms)", 2016, Berlin, Wiley-VCH Publishing House. Regarding the KI-based optimization unit, it should be pointed out again that other KI methods or concepts can also be used as the above-mentioned neural network. In particular, the training method can be modified and adapted to the corresponding requirements. In addition, the use of the indefinite article "one" or "one" does not exclude that the relevant features may exist in plurality. Likewise, the term "unit" does not exclude the fact that it is formed by a plurality of interacting subcomponents, which subcomponents may also be spatially distributed.

[0495] Reference numerals list

[0496] 1 Production equipment / laser melting equipment

[0497] 2. Product / Part / Object

[0498] 2'Products / Components / Buffers

[0499] 2” Products / Components / Square Rods

[0500] 3 Processing area / processing room

[0501] 4 Chamber wall

[0502] 5. Container

[0503] 6 container wall

[0504] 7 Working surface

[0505] 8 Tectonic Field

[0506] 10 carrier

[0507] 11 substrate

[0508] 12 Construction Platform

[0509] 13 Construction materials (in container 5)

[0510] 14 Storage

[0511] 15 Construction materials (in storage 14)

[0512] 16 Coating machine

[0513] 17 Radiant Heater

[0514] 20Irradiation device / exposure device

[0515] 21 Lasers

[0516] 22 Energy beam impact surface

[0517] 23 Deflection device / scanner

[0518] 24 Focusing device

[0519] 25 coupling window

[0520] 50 control devices

[0521] 51 Control Unit

[0522] 53 Irradiation control interface

[0523] 54, 54' Control data generating device

[0524] 55 bus

[0525] 56Terminal

[0526] 57Data Generation Unit

[0527] 58 Decision-making Unit

[0528] 60 means for generating optimized process variable values

[0529] 61 Request interface unit

[0530] 62 interfaces

[0531] 63 interfaces

[0532] 64 process variable interface unit

[0533] 65 Optimization Unit / Optimizer

[0534] 70 means for determining a property value

[0535] 71 Macroscopic property determination unit

[0536] 72 parameter set interface unit

[0537] 73 Scan direction interface unit

[0538] 74 interfaces

[0539] 75 interface

[0540] 80 Inspection device

[0541] AD Request Data

[0542] BEW basic property values

[0543] BSD Control Data / Exposure Control Data

[0544] DBS Attribute Database System / Database System

[0545] DS Data Storage

[0546] E Energy beam / laser beam

[0547] EDA Macro Property Database

[0548] EDB Basic Attribute Database

[0549] ERR Error value

[0550] ES Method Steps - Decision Making

[0551] FS Focus control data

[0552] G domain

[0553] GD geometry data

[0554] H Horizontal direction

[0555] HS Heating Control Data

[0556] HS2, HS3 layer scan direction arrangement / fill direction arrangement / fill strategy

[0557] HWR Main heat flow direction

[0558] KB back-propagation step

[0559] KNSP, KNWS combined neural network

[0560] KPS candidate parameter set

[0561] L, L1, L2, L3, L4 layer / layer

[0562] LH Hidden Layer

[0563] LI Input Layer

[0564] LO Output Layer

[0565] LS Laser Control Data

[0566] MS microstructure

[0567] MS1, MS2, MS3 Method Steps

[0568] MWA Macro Property Value

[0569] NN KI-based optimization unit / neural network

[0570] NNW, NPS, NSV KI-based optimization unit / neural network

[0571] NO, NO' Classic optimization process / optimizer

[0572] PAS Parameter Set Selector

[0573] PGO optimizes process variable values

[0574] PS Parameter Set

[0575] PS' Starting parameter set

[0576] PS ON , PSV ON , PSV' ON ,PSV” ON Predicted value

[0577] PS OR , PSV OR , PSV' OR ,PSV” OR Reference value

[0578] PSV” output variable

[0579] PSσ Input variables / stress optimization parameter set

[0580] Input variables / temperature optimization parameter sets

[0581] PSD Control Data / Process Control Data

[0582] PSS parameter set fitness value / PS score

[0583] PSSG Total PS score

[0584] QA Quality Requirements / Quality Requirements Data

[0585] RO Reference Orientation

[0586] S Scanning direction / moving direction of impact surface

[0587] SD Scan Control Data

[0588] SG segment

[0589] SG' Start segment

[0590] SGG Segment Boundary

[0591] SGI segment information

[0592] SG0 Powder segment

[0593] SG1, SG2, SG3

[0594] SSV segment scanning direction distribution

[0595] SSV' Start segment scanning direction distribution

[0596] SSV1, SSV2, SSV3, SSV4 segment scanning direction distribution

[0597] ST coating control data

[0598] S0 to S10 Method steps

[0599] S31 to S36 Sub-steps

[0600] Substeps S6a, S6b, S6c, S9a, S9b

[0601] TE Method Steps - End

[0602] TF1, ..., TFi, ..., TFn sub-function / sub-function / sub-function

[0603] TSD Carrier Control Data

[0604] TX Texture

[0605] V vertical direction

[0606] VG Method Steps - Comparison

[0607] Spatial direction in x,y plane

[0608] z Main structural direction

[0609] ZF objective function

[0610] φ α Process parameter set

[0611] Homogenized field quantities / homogenized stress states / input variables

[0612] σ 11 , σ 22 , σ 33 , σ 12 , σ 13 , σ 23 Matrix Elements for Homogenized Stress State

[0613] Temperature change function / cooling rate / input variables

[0614] ψ1,ψ2,……,ψ 360 Segment scan direction distribution SSV value.

Claims

1. A method for generating optimized process variable values ​​(PGO) for an additive construction process of a product (2, 2', 2") consisting of a multilayer (L, L1, L2, L3, L4) construction material (13), the method comprising the following method steps: - Provide the required data of the product (2, 2', 2") - Execute optimization methods to take into account the required data , wherein at least one optimized scanning direction distribution (SSV) for at least one area of ​​the raw product (2, 2', 2") is determined as the optimized process variable value (PGO) using a KI-based optimization unit (NN, NPS, NSV, NNW, KNSP, KNWS), - providing said optimized process variable value (PGO).

2. The method according to claim 1, wherein: In the optimization method, at least one best parameter set (PS) comprising a defined group of process parameter values ​​is selected from several candidate parameter sets (KPS) as at least one further optimized process variable value (PGO), preferably also using a KI-based optimization unit (NN, NPS, NSV, NNW, KNSP, KNWS).

3. The method according to claim 1 or 2, wherein: The product (2, 2', 2") is divided into a plurality of segments (SG, SG1, SG2, SG3) using the requirement data (AD), in particular the geometric data (GD) of the requirement data (AD), and the optimization method is performed in such a way that an optimized process variable value (PGO) is determined for each of the segments (SG, SG1, SG2, SG3) each time, preferably an optimal parameter set (PS) and an optimized segment scanning direction distribution (SSV) are determined for each segment (SG, SG1, SG2, SG3).

4. The method according to any one of the preceding claims, wherein: The optimization method comprises a plurality of iteration steps, and in at least one iteration step at least one KI-based optimization unit (NN, NPS, NSV, NNW, KNSP, KNWS) is used. and / or Therein, in the optimization method, at least one starting process variable value (PS', SSV') is first determined using a KI-based optimization unit (NN, NPS, NSV, NNW, KNSP, KNWS).

5. The method according to any one of the preceding claims, wherein: At least one KI-based optimization unit (NN, NPS, NSV, NNW, KNSP, KNWS) includes at least one neural network (NN, NPS, NSV, NNW, KNSP, KNWS).

6. The method according to any one of the preceding claims, wherein: determining at least one parameter set fitness value (PSS) for at least one area of ​​the product (2, 2', 2"), for at least several possible scanning direction distributions and / or for at least a part of the candidate parameter sets (KPS), respectively, and using the parameter set fitness value (PSS) to determine an optimized scanning direction distribution (SSV) and / or to select an optimal parameter set (PS) from the candidate parameter sets (KPS), Preferably, for different pairings of segment scanning direction distributions (SSV) and candidate parameter sets (KPS), parameter set fitness values ​​(PSS) are determined respectively.

7. The method according to claim 6, wherein for at least a portion of the scanning direction distribution (SSV) and / or the candidate parameter set (KPS), respectively determining Multiple requirements for specific parameter set fitness values ​​(PSS), in, Particularly preferably, the requirement-specific parameter set suitability values ​​(PSS) for the scanning direction profile (SSV) and / or the candidate parameter set (KPS) are each combined to form an overall parameter set suitability value (PSSG).

8. A method according to claim 7, wherein in the optimization method, an optimized process variable value (PGO) for the raw product (2, 2', 2") is determined in the following manner, namely: an optimized process variable value (PGO) is determined for each of the segments (SG, SG1, SG2, SG3), and the optimized process variable value is optimized within the corresponding segment for the total parameter set fitness value (PSSG) and in the raw product (2, 2', 2").

9. The method according to any one of claims 6 to 8, wherein: using said KI-based optimization unit (NN, NPS, NSV, NNW, KNSP, KNWS) to determine a scan direction distribution (SSV) for a segment (SG, SG1, SG2, SG3) and / or to select an optimal parameter set (PS) from said candidate parameter sets (KPS), Among them, preferably, a KI-based optimization unit (NN, NPS, NSV, NNW, KNSP, KNWS) is used, and when the KI-based optimization unit is generated, the parameter set fitness value (PSS) is used to perform training on the KI-based optimization unit (NN, NPS, NSV, NNW, KNSP, KNWS).

10. The method according to claim 8 , wherein a combined optimization method, preferably a heuristic approximation method, particularly preferably a simulated annealing method and / or a quantum annealing method is used to determine the process variable value (PGO) that is optimized for the sum of the fitness values ​​of the parameter sets in the product (2, 2′, 2″).

11. A method for generating control data (BSD, PSD) of a production device (1) for the additive construction of at least one product (2, 2', 2") consisting of a multilayer (L, L1, L2, L3, L4) construction material (13) in an additive construction process, the method comprising the following method steps: - the method according to any of the preceding claims provides for generating optimized process variable values ​​(PGO) for an additive construction process, - generating the control data (BSD, PSD) for the production device (1) in such a way that the optimized process variable value (PGO) is fully achieved in the additive structuring process according to a predefined evaluation criterion, in, The construction material (13) is preferably constructed and selectively solidified during the additive construction process, wherein, to achieve solidification, the construction material (13) is irradiated by at least one energy beam (E) on a construction field (8), wherein an impact surface (22) of the energy beam (E) moves on the construction field (8) so as to melt the construction material (13) in a target area within and around the impact surface (22).

12. A method for controlling a production device (1) for additively structuring a product (2), wherein: The method according to claim 11 generates control data (BSD, PSD) for the production device (1) and uses the control data (BSD, PSD) to control the production device (1).

13. A method for creating a KI-based optimization unit (KNSP, KNWS), in particular for use in a method according to one of claims 1 to 10 in order to provide a plurality of different types of requirement data. Determine the optimized process variable value (PGO), in, first - training at least one first KI-based optimization unit (NPS, NSV, NNW), said first KI-based optimization unit being based on a first type of requirement data to determine an optimized process variable value (PGO) for a process variable of a first type (PS), as well as - training at least one KI-based second optimization unit (NPS, NSV, NNW), said KI-based second optimization unit being based on the requirement data of said first type to determine an optimized process variable value (PGO) for a process variable of the second type (SSV), or the second optimization unit based on KI determines an optimized process variable value (PGO) for a process variable of the second type (SSV) based on the requirement data of the second type to determine an optimized process variable value (PGO) for a process variable of the first type (PS) and / or an optimized process variable value (PGO) for a process variable of the second type (SSV), And then a training method is used to create a KI-based combined optimization unit (KNSP, KNWS), in which at least the first KI-based optimization unit and the second KI-based optimization unit (NPS, NSV, NNW) are combined with each other to monitor the training of the KI-based combined optimization unit (KNSP, KNWS).

14. A device (60) for generating optimized process variable values ​​(PGO) of an additive construction process for producing a product (2, 2', 2"), comprising the following components: - a requirement interface unit (61) designed to provide requirement data of the product (2, 2', 2"); - an optimization unit (65) designed to take into account the requirement data performing an optimization method for determining the optimized process variable value (PGO) under the condition of: ... - a process variable value interface unit (64) designed to provide the optimized process variable value (PGO).

15. A control data generating device (54, 54') for generating control data (BSD, PSD) for a production device (1) for additively structuring a product (2) in an additive construction process, During the construction process, preferably the construction material (13) is constructed and selectively cured, wherein: To achieve solidification, the construction material (13) is irradiated by at least one energy beam (E) on a construction field (8), wherein an impact surface (22) of the energy beam (E) moves on the construction field (8) in order to melt the construction material (13) in a target area in and around the impact surface (22), Wherein, the control data generating device (54, 54') comprises at least the following components: - a device (60) according to claim 14 and / or an interface of a device (60) according to claim 14 for receiving an optimized process variable value (PGO), - a data generation unit for generating the control data (BSD, PSD) of the production device (1) in such a way that the optimized process variable values ​​(PGO) are adequately achieved in the additive construction process according to predefined evaluation criteria.

16. A control device (50) for a production device (1) for additively constructing a product (2) in an additive construction process, wherein: The control device (20) having a control data generating device (54) according to claim 15 for receiving control data (BSD, PSD) and / or an interface to a control data generating device (54') according to claim 15, And is designed to control the production device (1) using the control data (BSD, PSD).

17. A production device (1) for the additive construction of a product (2) in an additive construction process, having at least one control device (50) according to claim 16.

18. A computer program product comprising a computer program that can be directly loaded into a computer unit, in particular directly loaded into a storage device of a device (60) for generating optimized process variable values ​​(PGO), a control data generating device (54, 54') or a control device (50) for an additively constructed production device (1) for producing a product (2), and having program segments so that when the computer program is executed in the computer unit, all steps of the method according to any one of claims 1 to 13 are performed.

19. Optimized process variable values ​​(PGO) for an additive construction process for producing a product (2, 2', 2"), said optimized process variable values ​​(PGO) being generated by the method according to any one of claims 1 to 10.

20. Control data (BSD, PSD) of a production device (1) for additively structuring at least one product (2, 2', 2") in an additive structuring process, the control data (BSD, PSD) being generated according to the method of claim 11.

Images