Systems and methods
The system uses sensor data and neural networks to simulate and adapt additive layer manufacturing processes, addressing quality and stability issues in selective laser melting by detecting anomalies and optimizing parameters in real-time.
Patent Information
- Application Number
- JP2022559845
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2020-04-01
- Filing Date
- 2021-03-08
- Publication Date
- 2024-12-04
- Estimated Expiration
- 2041-03-08
AI Technical Summary
Existing additive layer manufacturing processes, particularly selective laser melting, face challenges in achieving consistent quality and stability of three-dimensional workpieces due to complex interactions and variables, making it difficult to ensure repeatable quality and efficiency, especially in large workpieces.
A system utilizing a sensor system with a data input device and processing device configured to generate a model for process monitoring, incorporating artificial neural networks to analyze sensor data and input parameters, allowing for in-situ simulation and adaptation of the manufacturing process to improve quality and efficiency.
The system enhances the quality and stability of three-dimensional workpieces by detecting anomalies and process errors in real-time, enabling adaptive adjustments to improve manufacturing outcomes and reduce defects.
Smart Images

Figure 0007594604000005 
Figure 0007594604000006 
Figure 0007594604000007
Abstract
Description
[Technical field]
[0001] The present invention generally relates to a system for simulating process (e.g., melt pool) monitoring by a sensor system (e.g., melt pool monitoring system) in an additive layer manufacturing process (particularly a selective laser melting process) used to solidify a material from which a three-dimensional workpiece is manufactured, a computer program product including program code portions for simulating process monitoring by a sensor system in an additive layer manufacturing process using a model generated using the system, a method for determining parameters influencing an additive layer manufacturing process used to solidify a material from which a three-dimensional workpiece is manufactured, a method for simulating process monitoring by a sensor system in an additive layer manufacturing process used to solidify a material from which a three-dimensional workpiece is manufactured, an apparatus for manufacturing a three-dimensional workpiece using an additive layer building method, a method for analyzing sensor data based on a comparison between sensor data and sensor data simulated using a model, a method for providing sensor data analyzed using the aforementioned method, a computer readable storage medium storing sensor data analyzed according to the aforementioned method, and a neural network for modeling an in-situ process (e.g., melt pool) monitoring system for additive layer manufacturing. [Background technology]
[0002] In additive layer methods, a workpiece is manufactured layer by layer by producing a series of consolidated interconnected workpiece layers. These processes can be distinguished by the type of raw material and / or the method by which said raw material is consolidated to produce the workpiece.
[0003] For example, powder bed fusion is a type of additive layer process that can process powdered, especially metallic and / or ceramic, raw materials into three-dimensional workpieces of complex shape. For this purpose, a raw material powder layer is applied onto a carrier and, depending on the desired shape of the workpiece to be produced, is exposed, for example, to electron or laser radiation in a site-selective manner. The radiation penetrating the powder layer causes heating and consequent melting or sintering of the raw material powder particles. Further raw material powder layers are then applied successively to the layer on the carrier already exposed to the radiation process until the workpiece has the desired shape and size. Selective electron beam melting, selective laser melting or laser sintering can be used, in particular, for the production of prototypes, tools, replacement parts or medical prostheses, such as dental or orthopedic prostheses, based on CAD data.
[0004] Throughout this disclosure, any reference to selective laser melting is equally applicable to selective laser sintering, selective electron beam melting, stereolithography, MELATO, selective heat sintering, or any other energy beam based additive processing method. Thus, any reference to additive layer manufacturing may be applicable to one or more of selective laser melting, selective laser sintering, selective electron beam melting, stereolithography, MELATO, selective heat sintering, and any other energy beam based additive processing method.
[0005] An important parameter of additive layer building methods is the quality of the manufactured workpieces. Furthermore, manufacturing efficiency is crucial, for example in the sense of keeping production cycles as short as possible. For example, numerous strategies are known to speed up the manufacturing of single workpiece layers. However, when manufacturing large workpieces, the known solutions do not always achieve the desired efficiency and / or quality.
[0006] Additive Manufacturing (AM) involves many complex variables that are monitored and controlled in the process to achieve acceptable quality of the printed part. A common approach to process monitoring is the measurement of melt pool properties. Analysis of properties can be complicated due to the amount of data collected, the unknown influence of process parameters and part geometry.
[0007] After more than 20 years of development, metal AM has established itself in industrial production environments for highly demanding markets, e.g. automotive, aerospace, or energy, based on increasingly better productivity, reproducibility, and reliability. In addition to all the known technical advantages, such as high geometric freedom and the possibility to create various geometric features, potential cost savings by partial redesign and application of lightweight construction methods, or higher part performance by optimized fluid properties or stress reduction, end-user industries need solutions how to document the entire build process and how to guarantee part quality, not only for highly stressed parts for safety-related applications. Selective laser melting involves a large number of complex variables, such as machine and environmental conditions as well as the placement, orientation, and process parameters of the build parts, making it difficult to guarantee stable, reliable conditions and to produce final parts with repeatable quality characteristics that reach the envisaged quality requirements.
[0008] A variety of quality assurance methods and techniques operating on-axis or off-axis are implemented or in development on state-of-the-art commercially available industrialized metal AM machines and new approaches are constantly being pursued in scientific environments to monitor and / or document irregularities that may affect part quality during a build job.
[0009] It is therefore an object of the present invention to improve the quality of three-dimensional workpieces, in particular those manufactured using an additive layer manufacturing process. It is further an object of the present invention to simulate an additive layer manufacturing process and thereby adapt the process to improve the quality of the three-dimensional workpiece. It is further an object of the present invention to determine parameters that affect the additive layer manufacturing process to a given extent. Summary of the Invention
[0010] Thus, the present invention describes a system for simulating process monitoring by a sensor system in an additive layer manufacturing process, in particular a selective laser melting process, used to solidify a material from which a three-dimensional workpiece is manufactured, comprising: a data input device (data input) that can be coupled to a data output device of the sensor system and configured to receive sensor data from the sensor system related to the process monitored by the sensor system, the data input device being further configured to receive one or more input parameters (i) related to the additive layer manufacturing process and / or (ii) related to an additive layer manufacturing device used for the additive layer manufacturing process and / or (iii) related to the process monitoring and / or the sensor system; and a processing device coupled to the data input device and configured to generate a model for simulating said process monitoring from the sensor data and the one or more input parameters.
[0011] The inventors recognized that while a common approach to process monitoring is in situ measurement (e.g., melt pool) characteristics using, for example, thermal radiation, analyzing the data, especially the characteristics without further knowledge of the system's behavior, can be complicated or even impossible due to the amount of data collected and the unknown effects of process parameters and part geometry for three-dimensional workpieces.
[0012] Implementations of the systems (and methods) described herein enable, inter alia, in-situ monitoring of a process (e.g., melt pool) and taking action during an additive layer manufacturing process (e.g., selective laser melting process) or even before a particular layer of the three-dimensional workpiece is manufactured to improve the quality of the three-dimensional workpiece. The system is thereby configured to simulate monitoring of the process (e.g., melt pool) in response to, inter alia, process parameters associated with the additive layer manufacturing process, and / or (ii) process parameters associated with an additive layer manufacturing apparatus used for the additive layer manufacturing process, and / or (iii) process parameters associated with a process monitoring and / or sensor system.
[0013] Throughout this disclosure, a melt pool may be defined as material (eg, which may originally be in the form of a powder, particularly a metal powder) that is heated during an additive layer manufacturing process to such an extent that the material melts into a liquid state.
[0014] The sensor system may include, for example, one or more cameras.
[0015] The implementations described herein may be used in addition to or as an alternative to the melt pool to monitor / simulate smoke and / or debris and / or splashes, whose trajectory and deposition position in the powder bed may be monitored / simulated during the exposure of one layer, which on the one hand can prevent direct irradiation and on the other hand can affect the exposure of subsequent layers through partial retention of splashes in the powder bed.
[0016] Additionally or alternatively, the monitoring / simulating may be in connection with the application of the layer (or images are taken after application of the layer and before exposure with the laser), in which, on the one hand, disturbances in the layer uniformity and deviations in the workpiece structure can be detected (if power is applied incorrectly, the solidified areas can, for example, be globally raised or show localized bumps (e.g. due to melt injection in a molten bath)) and, on the other hand, conclusions can be drawn about the just exposed layer and local layer thicknesses by means of the thermographic images.
[0017] In some example systems, the process monitoring includes monitoring of a melt pool, the sensor system includes a melt pool monitoring system, and the model for simulating said process monitoring includes a model for simulating said monitoring of the melt pool, and the sensor data related to the melt pool monitored by the melt pool monitoring system includes sensor data related to one or more of: characteristics of the melt pool, radiation emanating from the melt pool, particularly thermal or bremsstrahlung radiation emanating from the melt pool or the area surrounding the melt pool, or radiation emanating from the area surrounding the melt pool, radiation emanating from solidified material (which may be included in the area surrounding the melt pool or may be the same as the area after the melt pool), characteristics of a plasma generated from an additive layer manufacturing process used to generate the melt pool, the plasma resulting from material evaporated from the melt pool, characteristics of the plasma, radiation emanating from the plasma, particularly thermal radiation emanating from the plasma, and radiation reflected and / or diffracted from the melt pool and / or plasma, particularly laser radiation emanating from a laser device used to generate the melt pool. These may be taken into account when providing one or more input parameters to the data input device of the system such that the processing device of the system generates a model based on the sensor data and the one or more input parameters. Taking into account one or more of the following characteristics of the melt pool, radiation originating from the melt pool, particularly thermal radiation originating from the melt pool or radiation originating from the area surrounding or following the melt pool, radiation originating from the solidified material, characteristics of the plasma generated from the additive layer manufacturing process used to generate the melt pool, plasma originating from material evaporated from the melt pool, radiation originating from the plasma, particularly thermal radiation originating from the plasma, radiation reflected and / or diffracted from the melt pool and / or plasma, particularly laser radiation originating from the laser device used to generate the melt pool, may enable an improved model to be generated by the processing device, which may be taken into account when preparing the three-dimensional workpiece to improve the quality of the workpiece.
[0018] In some examples of the system, the one or more input parameters are related to or calculated based on the distribution and / or arrangement of the area and / or volume of unsolidified and solidified materials relative to each other in the detection range of the sensor system. Additionally or alternatively, the one or more input parameters are related to or calculated based on the ratio between unsolidified and solidified materials in the detection range of the sensor system, in particular the ratio between unsolidified and solidified materials in the detection range includes the ratio between unsolidified and solidified materials in the detection range in a predetermined number of material layers (one or more). Unsolidified materials within the meaning of the present disclosure may be defined as materials that are not irradiated by an energy / irradiation beam, although it is understood that the materials may exist in a solid state, i.e., as a powder, foil, or sheet, in some examples. In contrast, solidified materials may be defined in accordance with the present disclosure as materials that are irradiated by an energy / irradiation beam to melt the material and then solidify as part of the workpiece. Additionally or alternatively, the one or more input parameters include one or more parameters that define the location of the solidified or unsolidified material in the detection range of the sensor system. Properties such as thermal conductivity, radiation characteristics, and temperature of the detection area may be taken into account when generating the model, which may be determined by the apportionment, distribution, and arrangement of the solidified and unsolidified materials relative to each other. These parameters may differ between the solidified and unsolidified materials. This may allow the model to be improved to improve the quality of the manufactured three-dimensional workpiece, especially when the model-based simulation is used to adapt the additive layer manufacturing process.
[0019] Thus, in some examples of the system, the one or more input parameters include one or more of the thermal conductivity, the radiation characteristics, and the temperature of the unsolidified and solidified materials, respectively, in the detection range of the sensor system. Additionally or alternatively, the one or more input parameters include a temperature dependence exponent of the intensity of the two wavelength spectra.
[0020] In some examples, the system is configured to calculate one or more input parameters as a function of time, which may allow adapting the model to simulate in-situ process monitoring during an additive manufacturing process, thereby improving the model by taking into account the current situation, ultimately resulting in an improvement in the quality of the three-dimensional workpiece.
[0021] In some examples, the system is configured to calculate one or more input parameters for the prescribed material layer based on one or more output parameters output based on the simulation for one or more material layers underlying the prescribed material layer. Taking into account the simulation for one or more material layers underlying the prescribed material layer, the model may be improved for the prescribed material layer based on one or more properties associated with the simulated one or more material layers underlying the prescribed material layer. The properties of the prescribed material layer may depend, among other things, on the properties of the one or more material layers underlying the prescribed material layer.
[0022] In some examples, the system is configured to calculate one or more input parameters for the first plurality of material layers based on one or more output parameters output based on the simulation for a second plurality of material layers underlying the first plurality of material layers. Taking into account the simulation of the second plurality of material layers underlying the first plurality of material layers, the model of the first plurality of material layers can be improved based on one or more properties associated with the simulated second plurality of material layers. In particular, calculating one or more input parameters for the first plurality of material layers can further improve the efficiency for generating models for various material layers.
[0023] In some examples, the system is configured to calculate one or more input parameters for the material layer or layers based on one or more parameters indicative of when one or more underlying layers were or are being irradiated. This is particularly advantageous because the radiation characteristics of one or more underlying layers depend on the time that has elapsed since the underlying layer was irradiated and / or the duration for which the underlying layer was or is being irradiated. This allows for the generated model to be improved for the material layer to which the model is applied.
[0024] In some examples, the one or more parameters indicative of when the one or more underlying layers have been or are being irradiated are position-dependent parameters, which may allow for further refinement of the model based on these one or more position-dependent parameters.
[0025] In some examples of the system, the one or more input parameters include one or more of a radiation beam power (e.g., laser power), a scanning speed of a radiation beam used in the additive layer manufacturing process, a hatch distance used in the additive layer manufacturing process, and a material layer thickness of the material, which may improve the generated model since these input parameters may in some examples significantly affect the measurement signals for process monitoring.
[0026] In some examples, the one or more input parameters relate to or include one or more of: diameter of the radiation (e.g., laser) beam, heat dissipation during the additive layer manufacturing process, physical properties of the material to be solidified, optical surfaces of the additive layer manufacturing apparatus and / or scanning device and / or laser device and / or layer deposition device, and / or gas flow and / or temperature distribution within the additive layer manufacturing apparatus, and a scanning strategy used within the additive layer manufacturing process, in particular the position of the radiation beam on the material layer and / or the deflection angle of the radiation beam, and / or the vector angle and / or vector length of a vector for scanning the radiation beam across the material layer, and / or the scanning direction for scanning the radiation beam across the material layer relative to the direction of the gas flow.
[0027] In some examples of the system, the processing device includes an artificial neural network, and the generation of the model includes the steps of generating a data set including the sensor data and one or more input parameters, splitting the data set into a test data set and a training data set, selecting a neural network model for the artificial neural network, training the artificial neural network based on the neural network model using the training data set, and evaluating the neural network model using the test data set. Machine learning algorithms based on artificial neuronal networks (ANN) for simulating, analyzing, and evaluating data of the in-situ sensor (e.g., melt pool monitoring (MPM)) system used may allow for efficient analysis of process monitoring data, even for large amounts of data that may be collected.
[0028] In some examples of the system, the evaluation includes defining requirements for the artificial neural network, and if metrics associated with the artificial neural network do not meet the requirements, modifying the structure and / or topology of the artificial neural network until the requirements are met. The structure and / or topology of the artificial neural network may thereby be improved for a particular use case of the model applied to simulate monitoring of a process (e.g., a melt pool). Generating the model may include training the artificial neural network based on the modified structure and / or topology of the artificial neural network.
[0029] In some examples, the system is configured to simulate sensor data output by the sensor system based on the model and the one or more input parameters. The sensor data output by the sensor system may include sensor data used by the processing device to generate the model and / or sensor data output by the sensor system after the model is generated. The simulated sensor data can then be advantageously subsequently used to adapt the additive manufacturing process.
[0030] In some examples, the system is configured to compare sensor data output by the sensor system to simulated sensor data and detect anomalies and / or process errors during the additive layer manufacturing process based on the comparison. Additionally or alternatively, the anomalies and / or process errors may be determined based on the comparison even before the additive layer manufacturing process is performed.
[0031] In some examples, the system is configured to perform said comparison in-situ during the additive layer manufacturing process, which can advantageously enable adapting the additive layer manufacturing process in-situ when needed.
[0032] In some examples, the system is configured to provide a signal to the additive manufacturing device based on the comparison to adapt the current additive manufacturing process, in particular by stopping the current additive manufacturing process. This can allow, in particular, to adapt parameters for the additive manufacturing process so as to ultimately improve the quality of the manufactured three-dimensional workpiece. Depending on the measured parameters, for example, an automatic adjustment of the parameters or simply a suggestion for a change can be made. Additionally or alternatively, a partial cancellation of the building process is also possible, for example, individual components can be cancelled in the order of several components.
[0033] In some examples, the system is configured to generate a database including the generated models for subsequent use in the simulation. The generated models in the database can be updated to improve performance of the models for one or more use cases.
[0034] In some examples, the system is further configured to adapt process parameters for the additive layer manufacturing process and / or the shape of the three-dimensional workpiece manufactured using the additive layer manufacturing process based on the simulation model. This may advantageously allow, among other things, to intervene in the current process and / or change the expected shape of the three-dimensional workpiece and adapt the process and / or the shape, ultimately improving the quality of the workpiece. This may be, for example, an automatic adjustment of the shape, or simply a suggestion for a change. It is also possible that an optimized shape is proposed even before the start of manufacturing.
[0035] In some example systems, training of the artificial neural network is based on minimizing the mean absolute error between sensor data and simulation data associated with said simulation of process monitoring, thereby improving the quality of the three-dimensional workpiece.
[0036] In some examples, the training of the artificial neural network is based on multiple topologies of the artificial neural network, each of which differs from the other topologies by the number of hidden layers and / or the number of neurons per hidden layer. An optimal topology for the artificial neural network for a particular use case is discovered to improve the generation of a model used to simulate the monitoring of a process.
[0037] In some examples, the system is configured to train the topology of the artificial neural network until a minimum mean absolute error is reached, which may enable minimizing any anomalies in the three-dimensional workpiece and / or process errors when manufacturing the three-dimensional workpiece.
[0038] In some examples, the artificial neural network includes a multidimensional artificial neural network, which can be particularly advantageous for adequately addressing the shape of three-dimensional workpieces being manufactured.
[0039] We further describe a computer program product that includes a program code portion for simulating process (e.g., melt pool) monitoring by a sensor system (e.g., melt pool monitoring system) in an additive layer manufacturing process using a model generated using the system according to any one of the implementations described herein. The computer program product may be stored, inter alia, on a computer readable storage medium.
[0040] We further describe a method for determining parameters influencing an additive layer manufacturing process used to solidify a material from which a three-dimensional workpiece is manufactured, comprising the steps of determining an area and / or volume of one or more layers of material, determining which parts of the area and / or volume of the one or more layers of material have already been solidified for a particular time point, and determining parameters influencing the additive layer manufacturing process depending on the parts of the area and / or volume of the one or more layers of material that have been determined to be already solidified. The parameters may then be used when generating a model that is used, for example, to simulate the monitoring of the process. As outlined above, solidified and unsolidified materials may have different properties, in particular heat dissipation properties, which may therefore be taken into account when generating the model.
[0041] In some examples, the parameters are determined based on a first ratio between solidified and unsolidified material in a (two-dimensional) area and / or volume for a particular point in time. The first ratio may refer here to the ratio between the amount of solidified material and the amount of unsolidified material, respectively. Additionally or alternatively, in some examples, the parameters are determined based on a second ratio between solidified and unsolidified material in an area and / or volume for a particular point in time to produce a three-dimensional workpiece.
[0042] In some examples, the parameters are determined based on the location of the previously solidified material in the region and / or in the volume, which may take into account, for example, whether the location is near an edge region of the detection region of the sensor system or whether the location is more central in the detection region.
[0043] In some examples, the area and / or volume is greater than the area and / or volume, respectively, of a three-dimensional workpiece fabricated at a particular point in time using an additive layer manufacturing process, and in particular, greater than the area and / or volume of a melt pool.
[0044] We further describe a method for simulating process (e.g., melt pool) monitoring by a sensor system (e.g., melt pool monitoring system) in an additive layer manufacturing process used to solidify a material from which a three-dimensional workpiece is manufactured, comprising determining parameters using a method according to any one of the implementations described above, and providing the parameters to a model for simulating said process monitoring, whereby the model may be particularly accurate in view of the parameters used in generating the model, whereby the parameters are determined as a function of the area and / or volumetric portions of one or more material layers that have been determined to have already been solidified.
[0045] In some examples, the method further includes determining reference parameters using the determined parameters and input parameters, the input parameters including one or more of parameters related to an environment of an apparatus used for the additive layer manufacturing process, parameters related to physical properties of the material to be solidified, parameters related to process monitoring, parameters related to an apparatus used for the additive layer manufacturing process, and parameters related to the additive layer manufacturing process, and further includes interpreting the sensor signals output by the sensor system using the reference parameters. In some examples, the reference parameters are determined using an artificial neural network. Inputs to the artificial neural network may include one or more constants and / or one or more one-dimensional time series. This allows the artificial neural network to process large amounts of data.
[0046] In some examples, the sensor signals are used (i) to determine parameters affecting the additive manufacturing process and / or (ii) as the input parameters to an artificial neural network, and thus a loop can be performed to improve the artificial neural network for a use case and / or the model used to simulate the monitoring of the process.
[0047] We further describe a computer program product comprising program code portions for performing a method according to any one of the above implementations when the computer program product is executed on one or more computing devices. The computer program product may in particular be stored on a computer-readable storage medium.
[0048] We further describe as implementations described herein an apparatus for manufacturing a three-dimensional workpiece using additive layer build (i.e., fabrication) techniques, the apparatus including a carrier configured to receive material for manufacturing the three-dimensional workpiece, a material supply configured to supply material to a preceding material layer on the carrier and / or the carrier, a layer deposition mechanism for forming the supplied material into a material layer on the preceding material layer on the carrier and / or the carrier, a solidification device configured to solidify the material supplied to the carrier and / or the preceding material layer on the carrier to manufacture the three-dimensional workpiece, a gas supply configured to supply a shielding gas to a region of the material layer solidified by the solidification device, a process chamber including the gas supply and the solidification device, a monitoring system for monitoring the region where the material is solidified, and a system for simulating monitoring of the process according to any one of the implementations described herein. In some examples, the apparatus further includes a computer program product according to any one of the implementations described herein.
[0049] We further describe a method comprising the steps of providing sensor data relating to a process monitored or previously monitored by a sensor system during an additive layer manufacturing process, and providing a model for simulating monitoring of the process, wherein the model is (i) generated using a system according to any one of the implementations described herein, and / or (ii) provided through a computer program product according to any one of the implementations described herein, and / or (iii) generated using parameters determined according to any one of the implementations described herein, and further comprising the step of analyzing the sensor data based on a comparison between the sensor data and sensor data simulated using the model.
[0050] In some examples, the process-related sensor data monitored by the sensor system includes sensor data related to one or more of: characteristics of the melt pool, radiation resulting from the melt pool, particularly thermal radiation resulting from the melt pool, radiation resulting from solidified material, characteristics of the plasma generated from the additive manufacturing process used to generate the melt pool, where the plasma results from material evaporated from the melt pool, characteristics of the plasma, radiation resulting from the plasma, particularly thermal radiation resulting from the plasma, and radiation reflected and / or diffracted from the melt pool and / or plasma, particularly laser radiation generated from a laser device used to generate the melt pool.
[0051] We further describe a method that includes providing sensor data that has been analyzed using a method according to any one of the implementations described above.
[0052] We further describe a computer-readable storage medium that stores sensor data analyzed according to a method according to any one of the implementations described above.
[0053] These and other aspects of the invention will now be further described, by way of example only, with reference to the accompanying drawings, in which like reference numbers refer to like parts and in which: [Brief description of the drawings]
[0054] [Figure 1] 1 illustrates a schematic diagram of an apparatus for manufacturing a three-dimensional workpiece, according to some implementations described herein. [Diagram 2] 1 illustrates a model of a technical system as a black box according to some implementations described herein. [Diagram 3] 1 illustrates the basic structure of an artificial neuron, in accordance with some implementations described herein. [Figure 4] 1 illustrates a schematic diagram of part alignment and positioning using reference and feature cubes in accordance with certain implementations described herein. [Diagram 5] 1 illustrates a flow diagram of a method involving testing and evaluation concepts in accordance with some implementations described herein. [Figure 6] 1 illustrates a schematic diagram of a black-box model with identified input parameters, according to certain implementations described herein. [Figure 6A] 1 illustrates a schematic diagram of a black-box model with identified input parameters, according to certain implementations described herein. [Figure 6B] 1 illustrates a schematic diagram of a black-box model with identified input parameters, according to certain implementations described herein. [Figure 7] Measured sensor values (left) and simulated sensor values (right) are shown according to some implementations described herein. [Figure 8] 4 shows the mean absolute error of a simulation according to some implementations described herein. [Figure 9] 13 shows example MPM measurement signals and simulations according to some implementations described herein. [Figure 10]Illustrates online and offline training of models according to some implementations described herein. [Figure 11] 1 illustrates a network of models connected to a database, according to some implementations described herein. [Figure 12] 1 illustrates a powder bed, a workpiece, and a registration volume and region according to some implementations described herein. [Figure 13] 1 illustrates a schematic block diagram of an apparatus for manufacturing a three-dimensional workpiece, according to some implementations described herein. [Figure 14] According to some implementations described herein, a method for determining parameters affecting an additive manufacturing process is presented. [Figure 15] 1 illustrates a method for determining reference parameters using determined parameters and input parameters, according to some implementations described herein. [Figure 16] A method is presented for providing analyzed sensor data on a computer-readable storage medium in accordance with certain implementations described herein. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0055] One of the main challenges in industrial applications of selective laser melting processes (or additive layer manufacturing in general) is unstable component quality, process stability and reproducibility. The reasons for these are diverse and versatile interaction mechanisms and causal principles, which cannot always be clearly traced due to their complexity and which may also change during the course of the manufacturing process.
[0056] Therefore, sensors and monitoring systems are increasingly being used for quality assurance and documentation of the manufacturing process.
[0057] FIG. 1 illustrates a schematic diagram of an apparatus 100 for manufacturing a three-dimensional workpiece according to implementations described herein.
[0058] In this example, the apparatus 100 includes a melt pool monitoring instrument or system 102 for monitoring the melt pool, which in this example includes a first photodiode 104, a second photodiode 106, a beam splitter 110, and a mirror 108. The beam splitter 110 can split the incident radiation according to different wavelengths.
[0059] The apparatus 100 further includes a laser power monitoring instrument or system 130 for monitoring the laser power. In this example, the laser power monitoring instrument or system 130 includes a photodiode 132.
[0060] In this example, a laser beam with a wavelength of 1064 nm is generated using a laser device (laser) 112. The laser beam is collimated in this example using a collimator 114. The collimated laser beam passes through a 3D focusing device 116 before reaching a beam splitter 118, where the laser beam travels on the one hand towards a laser power monitoring device or system 130 and on the other hand through a scanner 120 towards the build chamber 122. Although not shown, it is clear that additional beam forming elements, i.e. additional focusing elements, or beam steering elements, i.e. additional scanning mirrors, may be present anywhere in the beam path. Furthermore, elements that do not affect the laser beam, such as an entry window to the build chamber, may be present.
[0061] The wavelength of radiation monitored by the melt pool monitoring instrument or system 102 is, in this example, greater than the laser beam wavelength, here 1064 nm. In other examples, wavelengths shorter than the laser beam wavelength can be used, or a spectrum including the laser wavelength can be used.
[0062] In this example, a powder application device 124 is provided for applying metal powder 126 onto a retractable platform 128 .
[0063] From the data stored via the melt pool monitoring system, it points out irregularities during the melting process and draws conclusions that may lead to potential defects in the components.
[0064] Depending on the amount of components and against the background of ever-increasing plant technology, one or more high-resolution process monitoring systems generate data volumes that are too large and too complex to be purposefully evaluated using manual and conventional methods of data processing.
[0065] Some of the implementations described herein also provide a data-driven approach that uses neural network-based machine learning algorithms to simulate an in-situ melt pool monitoring system and propose several use cases for the simulated in-situ melt pool monitoring system. Input features of the system model, in some examples, include information about process parameters and build strategies. Using these input features, the model can be trained to predict normal sensor outputs of the in-situ melt pool monitoring system. Comparison of stored data with simulated data can enable analysis to identify potential irregularities during fusion that may indicate anomalies during build based on a standard or are indicative of process instability.
[0066] Using a model of the melt pool monitoring or monitoring system, the simulated and measured sensor signals can be compared - a reference created specifically for the application, unknown in this way. Using this comparison, process variations or instabilities and / or any anomalies or component defects can be detected during the construction process, so that appropriate corrective measures can be taken to adjust the production parameters in real time, the components can be reworked or removed from further processing, or the knowledge database can be fed in parallel with the corresponding process events. Based on a well-established knowledge base, process variations or instabilities and / or any anomalies or component errors can also be taken into account before the production process, i.e. during the preparation of the production process, in particular during the pre-processing of the corresponding production parameters of the production process, i.e. during the placement, orientation and transport of a component or a number of components. This can increase the economic efficiency of the machine in terms of material consumption, production time or downstream post-processing procedures, as well as the quality of the manufactured components, and potentially no longer necessary destructive or non-destructive material inspections. In order to fairly evaluate the industry requirements of the “first-time-right” principle (i.e. the workpiece is manufactured with a constant quality from the very beginning of the manufacturing process), it may be advantageous to simulate the process. Using a simulation model of melt pool monitoring or a corresponding arrangement of additional or multiple monitoring systems, it may be possible even before the actual production process to optimize target parameters such as production time or quality as a function of process parameters or the geometry of the component / workpiece. This is done on the basis of in-situ process monitoring, which is used here in contrast to available simulation applications, e.g. thermal simulation, since the actual process data serves as the basis for modeling or simulation. Using a simulation model of MPM, the process parameters or the geometry of the part can already be optimized with respect to the thermal load before the actual building process.
[0067] The objective of the present disclosure is based on the identification of local anomalies as well as the identification of unspecified process stability. The definition of process stability may depend on given quality requirements. Considering the intensive study of signal characteristics as well as the system, both this general concept (method) and some of the applied input variables of the neural network can be developed using the system and method described herein.
[0068] If process monitoring systems from today indicate irregularities, most of the data or analysis is available after the entire build job (offline). The goal according to the implementations described herein is to identify condition or part irregularities before cost and time efficient post-processing non-destructive testing (NDT) methods, especially computer tomography (CT), have to be used on all final parts. There may be cases where reliable quality indicators are needed, which may be available during the process (online) in order to use further control approaches, to remove single parts of the job, or to have the ability to transfer and use this information for partial post-NDT processes. The present disclosure deals with the development, implementation, and evaluation, in particular of artificial neural networks (NNs), for modeling MPMs. For this purpose, it may be necessary to determine all relevant physical input parameters. With regard to the input and output parameters of the model, a suitable NN may need to be selected and a corresponding topology developed. The pre-processed data may be used to train and test the implemented NNs. The objective is to develop a sufficiently accurate input-output model that allows a comparison of simulated and measured sensor signals, independent of the geometry of the manufactured part / workpiece, which can then be used in the future to detect signal anomalies and process variations that can be correlated with quality characteristics of the component.
[0069] As will be understood by those skilled in the art, bid data refers to, for example, amounts of data that are too large, too complex, too fast moving, or too weakly structured to be evaluated using manual and traditional data processing methods. Artificial intelligence (AI) refers to all technologies used in connection with the provision of intelligent services previously reserved for humans. "Machine Learning" (MI), "Deep Learning" (DL), "Natural Language Processing" (NLP), and "Neural Networks" (NN) are corresponding subareas of AI and, in part, subareas within these subareas. Artificial neural networks (ANNs) have been used for scientific applications for decades. However, in recent years, significant advances in hardware for parallel computing and the availability of large data sets have led to an increase in the popularity of NNs for many applications in different domains. The architectural flexibility of a given NN allows it to be tailored to specific requirements and trained for many different applications. ANNs are mathematical models that simulate information processes in biological NNs. The mathematical mapping of neurons, and thus the development of ANNs, can be traced back to fundamental work known in the art.
[0070] In relation to system modeling with neural networks, modeling of technical systems generally aims to understand the causal principles of a real system and to express it mathematically. Such systems are also called black boxes, and only input and output variables may be used for their identification, as shown in Figure 2. In terms of artificial intelligence and machine learning, black box models are understood to be of a statistical nature, in contrast to white box models, which are based on analytical and physical descriptions and are therefore highly complex. Grey box models combine these approaches. All three models can be used for the subject of the present invention, but black box is the most preferred and white box is the least preferred.
[0071] The determination of a mathematical model of a system behavior from measurements of input and output variables is called experimental system identification. For complex, nonlinear and dynamic systems ANNs are suitable as universal approximations. An important feature of these networks is their adaptability. By training an ANN with the input and output variables of the real system, the internal parameters of the network are adapted to target the most accurate possible system picture. The structure of the ANN or topology may depend on the application and characteristics of the system to be modeled. ANNs can be divided into three classes: Multilayer Perceptron (MLP), Recurrent Neural Network (RNN) and Convolutional Neural Network (CNN).
[0072] Based on biological nerve cells, artificial neurons form the smallest processing unit of an ANN. The basic structure of an artificial neuron is shown in Figure 3.
[0073] A neuron y receives one or more inputs x i and one output y i It has weight w ij describes the strength of the information flow along the link. Depending on the sign of the weight, the inlet can act to inhibit or excite. If the weight is zero, the connection can be considered as not existing. The propagation function converts the weighted inputs into a so-called net input net j whereby this function usually only forms the sum of the inputs. Other theoretical correlations are possible. The inputs of a network of neurons can therefore be summarized in the following formula:
[0074]
number
[0075] The neuron output is formed from the network input by an activation function f, which is often a sigmoid function. Depending on the mesh topology, other nonlinear, piecewise linear or jump functions can be used. In general, the activation function is monotonically increasing. Thus, the neuron output can be described by the following equation:
[0076]
number
[0077] An ANN consists of at least one or more neurons in parallel. The type or topology of the NN determines how the different neurons are connected. The network consists of at least one input layer and one output layer, typically with a hidden layer between these two layers. The number of neurons in the input layer depends on the number of input parameters to be considered. Depending on the task, the number of neurons in the output layer can vary.
[0078] To reduce network error, gradient-based methods are often used that determine the direction in which the error decreases most quickly, similar to finding a minimum in a system of nonlinear equations such as a cost function. The goal of training is to minimize the cost function without overfitting the ANN, so that the ANN does not memorize the training data set and normal variations in the cost function, such as noise, are avoided. To solve this problem, the training and test sets are used to set an optimal cost function between variance and bias.
[0079] Currently, there are several approaches in the development stage where AI can be used for error detection during process monitoring using artificial intelligence for additive manufacturing. Most of these approaches differ, among other things, in the sensors used. Images may be taken before and after exposure with a high-resolution reflex camera. After the build process has been carried out, the part is measured using CT and the errors found serve as a reference on which a supervised machine learning algorithm can then be implemented that correlates the acquired image data with the CT measurements. The trained algorithm can then detect defects in the part using only the image data acquired during the process. Alternatively, the shape of the melt pool is stored with a high-speed camera limited to a small area in the working surface. Alternatively, the thermal radiation of the melt pool or a combination of the thermal radiation by a pyrometer coaxially and the dynamics of the melt pool by a high-speed camera are analyzed and correlated with defects or mechanical properties of the manufactured component.
[0080] The inventors recognized that most of the existing work has focused on defect detection and location in simple cube geometries and can only be transferred to a limited extent to other more complex component geometries, and that the existing sensors used often cannot be integrated into commercially available metal AM machines.
[0081] Some of the investigations presented in this work were carried out using a SLM 280 with a focal diameter of 80 pm to 115 pm and a laser of 700 W, equipped with an MPM process monitoring system - a coaxial in-situ measurement instrument based on two photodiodes and an FPGA operating at 100 kHz and 16-bit x / y coordinate presentation. HL This was done with a single optical instrument. In the following approach, the sensor data for photodiode 1 corresponds to channel 1, photodiode 2 corresponds to channel 2, etc.
[0082] Further details relating to the setup of the mechanical and optical systems as well as the hardware, software and measurement principle of the MPM system can be found in D. Alberts, D. Schwarze and G. Witt, “High speed melt pool and laser power monitoring for selective laser melt-ing(SLM®)” 9th International Conference on Photonic Technologies, LANE, 2016, https: / / www.lane-confer-ence.org / app / download / 11537198949 / LANE2016_1219_Alberts_IC_endformat.pdf?t=1534490222 and D. Alberts, D. Schwarze and G. Witt, “In situ melt pool monitoring and the correlation to part density of In-conel®718 for quality assurance in selective laser melting”, Solid Freeform Fabrication: Proceedings of the 28th Annual International, 2017, pp. 1481-1495.
[0083] Official documentation files developed and verified at the time of IN718 from SLM Solutions Group AG® are used. Tests result in an average part density of 99.8% so that the manufactured parts can be used as a reference with almost no defects. Boundary conditions such as, for example, inert gas atmosphere (argon), oxygen content (<0.1%), gas flow rate, preheat temperature (200 °C) or pressure ratio are monitored throughout the process and are manufactured under constant conditions with the help of an integrated control system. The powder material used consists of spherical shapes with a particle size distribution of 10-45 pm.
[0084] For this approach, two build jobs were manufactured: one with five parts and reference parameters based on a basic cube shape (reference cube), and one with five parts and reference parameters containing features of different shapes (feature cubes). The arrangement and positioning of the parts are shown in Figure 4.
[0085] Shape features include horizontal, cylindrical holes with various diameters and rectangles with various lengths, as well as vertical, thin pointed walls or powder inclusions and triangles with different overhangs and opening angles. For each layer, a single portion exposure is made from left to right against the gas flow. As a hatch pattern, a bidirectional stripe exposure and an increase in angle after each layer is used.
[0086] FIG. 5 shows a flow diagram of a method including the concept of testing and evaluation.
[0087] For each layer manufactured, the MPM generates, for example, a binary file containing the measurements for each measurement cycle and the process data of the build process, from which all input parameters relevant to the ANN are then calculated.
[0088] A data set (1 to n binary data) consisting of input data (process data and calculated input parameters) and output data (sensor data from channel 1 and channel 2) forms the basis for training and testing the ANN.
[0089] The compiled dataset is pre-processed (data with different formats may be converted to have the same format, values may be scaled, boundary values may be defined) and split into a test dataset and a training dataset.
[0090] During performance evaluation, a test data set is used to test the trained network using data that was not used for training.
[0091] For performance evaluation, appropriate metrics are used to evaluate the ANN. If these metrics do not meet the requirements of the system model, it may be necessary to optimize the network structure or topology. The network with the modified topology may then need to be trained and evaluated again. These steps may be repeated until a network topology is found that is capable of mapping the system behavior of the MPM in the best possible way.
[0092] Subsequently, in this example, the system model is evaluated for its suitability as a quality assurance tool in the AM process of each metal of selective laser melting (in the broadest sense since signal anomalies or process errors are previously unknown) by using the feature cube dataset. The system model simulates the sensor output based on the input data of the dataset used.
[0093] The simulated sensor outputs may then be compared to the actual sensor outputs. The system model is then evaluated for its ability to detect or indicate any signal anomalies and / or process errors.
[0094] In another approach, when shape dependencies are determined, the system model can be extended by training the existing one using the dataset from the feature cube build process and evaluating the performance.
[0095] When developing a system model, the entire system, including the architecture or functionality of the MPM and existing databases, may need to be analyzed. Based on this, physically relevant input parameters are identified, a suitable network model is selected for the system model (e.g., based on requirements related to accuracy), and requirements for this network model are specified. Only in-process parameters are taken into account. Neither the metal powder as well as the process steps upstream and downstream of the process are further considered in this example. The selective laser melting process and the MPM are assumed to be black boxes, and only input and output parameters are considered. The objective of the analysis is to characterize the important influencing parameters. Statistically significant parameters may be parameters whose effect is greater than random scattering.
[0096] The melt pool emits thermal radiation according to its temperature, area and position. In this example, the detection ranges of the two photodiodes can be assumed to be identical and circular. The radiation intensity measured at the detection ranges is interpolated and averaged over a measurement period of, for example, 10 μs and stored with the current position, which represents the measurement area at that time. The process parameters, and therefore the temperature and size of the melt pool, are approximately constant (in some examples), but the measured intensity of the vectors can vary.
[0097] In addition to these intensity fluctuations, there is usually higher frequency noise. To qualitatively estimate the noise in the construction process used, the sensor values of the build jobs used can be smoothed by a moving average value.
[0098] The relevant input parameters correspond to the influencing factors, which are significantly influenced by the measurement signal of the MPM. The process parameters with laser power, hatch distance, layer thickness, and scanning speed can affect both the part quality and the measurement data, which may be relevant. Fundamental differences between the materials used and the machine-specific conditions may need to be taken into account. Both the negative effects related to the causes on the part quality and the causes summarized here can be revealed. A changed beam diameter can mean an enlargement of the melt pool and therefore need to be taken into account. Another effect is the incidence angle of the laser on the powder bed, taking into account the scanning direction. In some cases, there may be a slight correlation between the interaction of the laser radiation with the metal vapor and the part quality. The scanning direction relative to the gas flow can affect the measurement signal. Exposure of the vector in the flow direction means that the thermal radiation of the melt pool is shielded by the welding fume, resulting in a reduction of the measurement signal.
[0099] In addition, it may be necessary to map properties such as thermal conductivity, radiation characteristics and temperature of the detection area, taking into account the measurement and shape dependencies determined by the allocation, distribution and arrangement of the solidified and powder materials relative to one another, the scanning strategy and the process parameters. The powder surrounding the part acts as a thermal insulator and has a lower thermal conductivity than the molten and solidified metals. The solidified and powder materials have different luminescence properties, so it is important to take their ratio into account in the detection range. The temperature of the already solidified material depends on a number of influences, and the thermal balance of the layer is characterized by the time sequence of the individual scanning vectors, the process parameters, the temperature control of the build chamber and the heat dissipation.
[0100] In addition to convection and thermal radiation, heat dissipation is determined by the thermal conduction of the apportionment, distribution and arrangement of the solidified and powder materials relative to one another, which will be considered as a geometry-dependent input parameter.
[0101] FIG. 6 shows an overview of the identified input parameters categorized into various groups including environment, human influenced parameters, materials, melt pool measurements, machine specific conditions, process parameters, scanning strategy, and geometry dependency, and illustrates the procedure of experimental modeling on a black-box system using an artificial neural network 602.
[0102] The input classes, subclasses and parameters represent possible combinations of various weighted influence variables on the quality of the represented system model (here thermal). Differentiation is only useful for structured representations, reducing the value of individual input parameters.
[0103] The input class Environmental describes the totality of the selection of environmental factors that occur during the manufacturing process, here referring to both stock and process periphery.
[0104] The input class human describes the totality of the selection of human factors that occurs mainly during the preparation of the manufacturing process.
[0105] The input class Material describes the entire selection of material parameters that occur before, during and after the building process, in this example focusing on its composition as well as its physical, chemical or optical properties.
[0106] The input class Measurement describes the totality of measurement influence selection that occurs before, during and after the build process, focusing on the methods, measurement equipment capabilities, accuracy, conditions, calibration and data acquisition.
[0107] The input class machine-specific conditions describes the entire selection of machine-specific influencing variables occurring before, during and after the build process and mainly relates to the conditions and characteristics of optical interfaces, scanning devices, gas flows, temperature distribution, behavior and conditions associated with powder application as quality parameters of the laser device used.
[0108] The input class method describes the entire selection of machine-specific influencing variables that occur before, during and after the build process. In this example, the emphasis is on process parameters such as laser power or deflection speed, scanning strategies such as position data or vector length, or process input parameters with shape-specific dependencies. These are characterized by spatial and time-dependent and partly temperature-dependent conditions or outcomes with or without laser-powder interaction, with respect to powder / solid material ratio.
[0109] In conjunction with the selection of the network model, the ANN can simulate output parameters based on given input parameters. To simulate this relationship between cause and effect, the ANN may need to be trained. Since the output data for training is already available, the present invention refers to supervised learning. The problem at hand can be described as a regression problem, since the initial data are continuous variables. The input and output data are available as one-dimensional time series.
[0110] When evaluating the performance of an ANN, in this example, the ANN is TIFF0007594604000003.tif6163 and the actual sensor output y i It is trained to minimize a cost function that measures the mean absolute error (MAE) between
[0111]
number
[0112] Measurement data of a process "reference cube" (Figure 4) consisting of 300 layers is used in this example for evaluation. 75% of this dataset is used for training and 25% for testing the NN. This dataset is used to train and test different NN topologies, differing in the number of hidden layers and the number of neurons per hidden layer. Using the training dataset, the topology is trained as long as it does not reach the minimum of the cost function for both channels. With the amount of epochs investigated, overfitting cannot occur. After each epoch, the topology is evaluated using the test dataset and the average absolute deviation is calculated.
[0113] As the size and complexity of the NN increases, the error made by the network in a simulation decreases. Increasing the complexity also increases the time required to train the network.
[0114] To illustrate the simulation accuracy, FIG. 7 shows actual measurements (left) and simulations (right) for channel 1 of layer 200 from the process "reference cube."
[0115] For the system model to be universally applicable for different components, the simulation may need to be independent of the part's geometry. The previously used system model "reference cube" is applied to the data of the process "feature cube" (Figure 4) and performance is evaluated. The existing special geometry cases are unknown to the "reference cube" system model and can therefore be considered as process instabilities (comparison of two identical series of processes) and / or process anomalies (defects in known geometries). The existing system model "reference cube" is then trained using components from the "feature cube" process to determine whether this can improve performance. The measurement data of the process "feature cube" consists of 668 layers. The resulting system model "feature cube" can then be compared to the original system model "reference cube". For evaluation, each layer of the part from the "feature cube" process is simulated with both system models and the MAE is calculated and analyzed. Figure 8 shows the results of the "feature cube" part using channel 1 as an example.
[0116] The performance of both system models is comparable for many layers. Layers 1-85 are the support structure, whose high deflection can be explained by the damped heat conduction downwards to the support, which will not be discussed further here. For the following layers, both system models have similar simulation accuracy, since the exposed areas of these layers are similar to those of the "reference cube" process. As soon as the layers change from the exposed areas for the special shape cases (layers 125-450 and layers 600-650), the MAE of the "reference cube" system model is getting worse (see Figure 8). Compared to the trained system model "reference cube", the trained system model "feature cube" shows almost the same simulation error for both channels.
[0117] 9 illustrates an example of different system behaviors for channel 1. It shows the MPM measurement signal (left), a simulation using the system model "reference cube" (center), and a simulation using the system model "feature cube" (right) for layers 175 (top), 521 (center), and 625 (bottom) of channel 1.
[0118] We visualize that the simulation using the system model "reference cube" shows large deviations, especially in the corners of the thin exposed areas and inner parts, while the system model "feature cube" shows smaller deviations from the MPM signal. It is also possible to simulate subregions using special cases of real shapes by training with the selected NN. Figure 8 shows that training brings the MAE of these layers back to the process noise range of about 160 counts.
[0119] As outlined above, by using initial base shapes as well as introducing shape features (simulation of real component shapes), model accuracy can be shown to be >99%, and at least >93% when taking into account the current overall noise of the system. It is therefore possible to train individual customer-specific reference models with a certain accuracy based on the same build job or only a few layers of this build job.
[0120] The model can be adapted as needed. What is shown is universal applicability to shapes, materials, or the like.
[0121] As well as offline training plus online training, offline evaluation plus online evaluation is also contemplated.
[0122] The online evaluation can then be used to derive decision trees such as pausing, stopping, closed-loop and open-loop control circuits, such as melt pool control by laser power, scanning speed, etc., or even adaptation of scanning strategies in the current or subsequent process. For example, partial cancellation of the build process may be possible, where parts assumed to be defective are cancelled, while non-defective parts of the build job are terminated. The simulation can be adapted online to parameter changes for partial cancellation.
[0123] Based on the methods and systems described herein, the shape, the component locations and component orientations on the building platform, as well as the individual topology optimization of the parameters used can be adapted, modified, and optimized by prior parameter calculations based on selective laser melting data files. Upstream conventional simulations may not be necessary, but may be possible.
[0124] The iterative process of determining and interpreting material process parameters as a function of material geometry can be time-saving, and various scenarios can be simulated in advance before the final build process involving any destructive or non-destructive testing is performed.
[0125] To further increase the model accuracy, the addition of further input variables may be considered, in particular whereby the position - i.e. the separation of consolidated and unconsolidated (powder) material - is taken into account by considering the detection area at time t.
[0126] Additionally or alternatively, in parallel with an already trained model, the input and output variables of the artificial neural network may be taken into account, taking into account the already generated layer n-1 or multiples thereof. In some examples, the third dimension (z-direction) may be taken into account when generating the model.
[0127] Considering previous output parameters as iterative input parameters to generate the model, the simulation can be more accurate compared to the (ideal) offline system model. This is shown in Figure 10, where an online and offline sensor model is generated, whereby in this example, the online sensor model output is fed back to improve the model.
[0128] In the SLM process, the local and global thermal balance is also influenced by the additional time and layer-by-layer time variables in the z-direction in terms of heating time and cooling rate, which are finally reflected in the output parameters.
[0129] The previously determined parameters can be used for further analytical approaches in addition to an initial target / actual comparison during the manufacturing process (online) as well as for training the model offline. Providing data for training during manufacturing time or at least with a slight delay is also conceivable.
[0130] In FIG. 11, a network of models connected to a database is shown. The above network topology can again be supplemented here as an input parameter for an extended optimization of the local - for example, at a lower level where isolated application systems of machines, materials of the same machine, a certain number of machine systems of a production plant or the like occur - network topology, to the extent that the resulting target / actual comparison of the complete production process can be performed based on the already existing network topology depending on the application. At one higher level - the top-level concept shown in the center of FIG. 11 - the network topology of the higher instance can also be used to disproportionately benefit from the provided data and to serve as a master system. In this case, a certain number of machine systems of manufacturers in different locations or a higher level of service of, for example, a machine manufacturer can be considered. All local concepts can benefit from this in that the advantage of knowledge can be made available to the local network based on significantly more usage data, either equally or filtered.
[0131] Several artificial neural networks can be combined. Existing networks can be implemented by multidimensional neural folding networks. This can allow better results for shape conditions in the z direction.
[0132] Additional measurement systems can be used to verify possible local defects.
[0133] The present disclosure shows that and how the data of the in-situ melt pool monitoring system used here can be simulated using a machine learning algorithm, in particular based on an artificial neural network. This ANN is able to predict the actual sensor output of the MPM system largely independent of the geometry. As the complexity of the part's geometry increases, the accuracy decreases. A comparison between measured and simulated data shows that by training with the NN, both part regions can be simulated with real geometry, and that process instabilities can be detected and used as an indicator of process stability. The results show that the simulation of MPM data can be used in a targeted manner and for the documentation and quality assurance of the respective metal additive manufacturing process of selective laser melting, depending on the task at hand.
[0134] Evaluation of the individual layers shows that for some complex geometries the deviation between measured and simulated signals is larger. On the one hand, this challenge can be solved by considering the allocation, distribution and arrangement of the solidified and unconsolidated (powder) materials relative to each other in the total build volume for the ANN. On the other hand, this finding suggests the possibility to derive and validate potentially further input parameters from the existing database, resulting in an ANN trained on the basis of a base geometry that is sufficient to identify further special cases and use cases. The simulation accuracy of complex geometries and the deviation between measured and simulated signals could potentially be improved by combining the ANN used with other network topologies. It may then occur that in addition to the current state, it is possible to move to longer process times using different materials as well as to more voluminous or higher parts that are entirely real. The comparison of measured and simulated data can be used for the quantitative detection of certain anomalies and their classification or for the detection of further process irregularities due to different error scenarios during the process. In this context, the total noise specific to the detector can be improved in order to then take appropriate actions to reduce it. Additionally, process parameters and part geometries can be optimized using MPM's simulation models with respect to thermal loads before the actual build process begins.
[0135] The system and method according to the present disclosure differs from the prior art in that data from post-processing monitoring such as CT data may not need to be used as a reference. Instead, it is based on the system behavior of the process monitoring using reference process parameters (user-specific process parameters defined as standard data from pre-processing in the SLM process such as component orientation and component positioning as well as laser power, or resulting heat dissipation as well as geometric features such as component overhangs, drill holes, and percentage of solidified material or percentage of powder material at time t in the detection area). With these process parameters, fewer or no defects occur (compared to when these parameters are not taken into account). If there is a deviation from this referenced system behavior, the component quality is not guaranteed and errors in the component structure may occur.
[0136] On the other hand, the systems and methods according to the present disclosure cannot be compared to "conventional" and established thermal simulations, as their underlying thermal / physical principles do not take into account the technological and process-related boundary conditions of powder bed based additive manufacturing processes using laser-metal powder interaction.
[0137] 12, a powder bed 1302 is shown having consolidated material 1308 and material to be consolidated 1306. Shown are different workpiece layers 1304. The surrounding powder bed 1302 is filled with unconsolidated material, which is not shown in the figure for clarity.
[0138] In this example, a registration volume 1312 is determined, which is two layers deep. Additionally, in this example, a registration area 1310 is determined, which is identical to the top surface of the registration volume 1312 and is at the top surface of the work plane, i.e., the powder bed 1302. In this example, the area and volume are larger than the area and volume in which the three-dimensional workpiece is manufactured at a particular time, in other words, the area and volume of the melt pool 1314. In the illustrated example, the melt pool 1314 is at the center of the registration area 1310. It is clear that in alternative embodiments, the melt pool 1314 can also be located inside the registration area 1310, or outside the registration area 1310, in particular eccentrically in front of the registration area 1310 when viewed from the movement direction. In this example, only one layer deep is shown, but it is apparent that the melt pool 1314 may be several layers deep and may even be deeper than the registration volume 1312, and thus the registration volume 1312 will only include a portion of the melt pool 1314.
[0139] For a particular time, it is determined which portions of the registration area 1310 and / or registration volume 1312 are already solidified and which portions of the portions are not yet solidified. Alternatively or additionally, it is determined which portions of the registration area 1310 and / or registration volume 1312 are already solidified and which portions are not yet solidified. Based on this determination, parameters influencing the additive layer manufacturing process are determined depending on which portions of the registration area 1310 and registration volume 1312 are already solidified and which portions of the portions are not yet solidified (i.e., their ratio and / or allocation and / or distribution and / or arrangement relative to each other). This determination is based on the fact that the powder surrounding the already solidified portions acts as an insulator and has a lower thermal conductivity than the molten or solidified material. For the determination of parameters influencing the additive layer manufacturing process, the processing time from the parts of the registration area 1310 and / or registration volume 1312 that are already solidified can also be taken into account (additionally or alternatively), and thus it is possible to determine the specific temperature of the parts of the registration area 1310 and / or registration volume 1312 that are already solidified at a specific time. This can also be used for 2D or 3D mapping. For the determination of the temperature or corresponding indicators of the already solidified parts, especially the lower layers, data can be used from simulations and / or actual sensor data from a melt pool monitoring system.
[0140] FIG. 13 shows an apparatus 1400 for manufacturing a three-dimensional workpiece.
[0141] In this example, the apparatus 1400 includes a simulation system 1402 including a processing unit 1404, a data input 1406, and an artificial neural network 1408 according to implementations described herein.
[0142] Further, in this example, the apparatus 1400 includes a melt pool monitoring system 1410 including one or more sensors 1414 for monitoring the melt pool during the selective laser melting process and a data output device 1412 for outputting data related to the monitoring of the melt pool.
[0143] In this example, the apparatus 1400 further includes a carrier 1416 configured to receive a powder material from which the three-dimensional object is manufactured. A layer deposition mechanism 1420 is provided to spread material supplied through the material supply device 1418 onto the carrier 1416.
[0144] In this example, the apparatus 1400 further includes a process chamber 1422 that includes a solidification device 1424 and a gas supply 1426 for providing process and / or shielding gases during the selective laser melting process.
[0145] In this example, the apparatus 1400 further includes a computer-readable storage medium 1428 for storing a model for simulating the melt pool monitored and / or analyzed / simulated sensor data, in particular in accordance with the implementations described herein.
[0146] FIG. 14 illustrates a method 1500 for determining parameters affecting an additive layer manufacturing process according to implementations described herein.
[0147] In this example, method 1500 includes, in step S1502, determining an area and / or volume of one or more layers of material. In step S1504, method 1500 includes determining which portions of the area and / or volume of one or more layers of material have already been solidified for a particular point in time. Then, in step S1506, parameters affecting the additive layer manufacturing process are determined as a function of the portions of the area and / or volume of one or more layers of material determined to have already been solidified.
[0148] FIG. 15 illustrates a method 1600 for determining reference parameters using determined parameters and input parameters according to implementations described herein.
[0149] In this example, method 1600 includes determining parameters that affect an additive layer manufacturing process used to solidify a material from which a three-dimensional workpiece is manufactured according to method 1500 shown in Figure 14. In step S1602, the parameters are provided to a model for simulating melt pool monitoring. Then, in step 1604, reference parameters are determined using the determined parameters and the input parameters.
[0150] FIG. 16 illustrates a method 1700 for providing analyzed sensor data on a computer-readable storage medium according to implementations described herein.
[0151] In this example, in step S1702, sensor data related to the melt pool monitored or pre-monitored by the melt pool monitoring system during the selective laser melting process is provided.
[0152] Further, in step S1704, a model is provided for simulating monitoring of the melt pool, which may be generated based on the melt pool of the current selective laser melting process and / or previously generated melt pools.
[0153] In step S1706, the method 1700 includes analyzing the sensor data based on a comparison between the sensor data and the sensor data simulated using the model.
[0154] Next, in step S1708, the analyzed sensor data is provided on a computer-readable storage medium.
[0155] We have further described a method for determining influence values for generating a three-dimensional structure through layer-by-layer integration of flowable (legged) material by energy input into the material at a particular location of the structure to be generated (or a virtual model thereof) at a given time point when the structure is created (or simulated), thereby providing a registration area or registration volume that is in a certain positional relationship to the particular location, the influence values being determined depending on the already created structure in the registration area / volume at a given time point.
[0156] The method comprises in particular - the specified location is the location of the energy input at the specified time, and / or the impact value is determined according to the ratio of structures already created at a given time to the remaining area / volume in the storage area / volume, and / or the influence value is determined as a function of the ratio of already generated structures at a particular time to not yet generated structures in the registration area / volume, and / or the influence value is determined as a function of the position of the energy input in the registration area / volume at a time already in the past, and / or the influence values are used for the interpretation of the measured values whose storage range corresponds to the measurement range, and / or the registration area is larger than the location of the energy input (and / or its interaction zone); and / or the storage volume has a depth of 1, 2 or more layer thicknesses, and / or the influence values are used to calculate a virtual simulation, and / or - on the basis of the simulation, a structural optimization is proposed; and / or the influence value serves as an input parameter for the determination of an effect value, the influence value being determined in dependence on at least one further input parameter selected from one or more of the following: one or more environmental parameters, one or more construction parameters, one or more material parameters, one or more measurement method parameters, one or more device parameters and one or more process parameters; and / or the effect value is determined as an output value of an artificial neural network, and / or one or more input parameters of the ANN are preferably constants or one-dimensional time series, and / or the actual value of the ANN is used as the reference value for the interpretation of the measurements, and / or - The measurements themselves determine the influence values and / or serve as input parameters for the ANN.
[0157] We further describe a computer program product for performing the aforementioned determinations.
[0158] No doubt many other effective alternatives will occur to those skilled in the art, and it will be apparent to those skilled in the art that the present invention is not limited to the described embodiments and implementations, and will be understood to encompass modifications within the scope of the claims appended hereto. [Configuration 1] A system for simulating process monitoring by a sensor system in an additive layer manufacturing process, in particular a selective laser melting process, used to solidify a material from which a three-dimensional workpiece is manufactured, comprising: a data input device coupleable to a data output device of the sensor system and configured to receive sensor data from the sensor system relating to the process monitored by the sensor system, the data input device comprising: (i) in connection with said additive manufacturing process; and / or (ii) relating to an additive manufacturing apparatus used for said additive manufacturing process; and / or (iii) associated with the process monitoring and / or the sensor system; a data input device further configured to receive one or more input parameters; a processing device coupled to the data input device and configured to generate a model from the sensor data and the one or more input parameters for simulating the process monitoring. [Configuration 2] the process monitoring includes monitoring a melt pool; the sensor system includes a melt pool monitoring system; the model for simulating the monitoring of the process includes a model for simulating the monitoring of the melt pool; The sensor data related to the melt pool monitored by the melt pool monitoring system includes: characteristics of the melt pool; radiation emanating from the melt pool, in particular thermal radiation emanating from the melt pool; radiation arising from solidified materials, a characteristic of a plasma generated from the additive manufacturing process used to generate the melt pool, the plasma resulting from material evaporated from the melt pool; radiation originating from the plasma, in particular thermal radiation originating from the plasma, and radiation reflected and / or diffracted from the melt pool and / or the plasma, in particular laser radiation originating from a laser device used to generate the melt pool; 2. The system of claim 1, further comprising sensor data relating to one or more of: [Configuration 3] 3. The system of configuration 1 or 2, wherein the one or more input parameters are related to or calculated based on the distribution and / or arrangement of areas and / or volumes of unsolidified and solidified material relative to one another in a detection range of the sensor system. [Configuration 4] 4. The system of any of configurations 1-3, wherein the one or more input parameters are related to or calculated based on a ratio between unsolidified and solidified material in the detection range of the sensor system. [Configuration 5] 5. The system of configuration 4, wherein the ratio between unsolidified material and solidified material in the detection range comprises a ratio between unsolidified material and solidified material in the detection range for a predetermined number of material layers. [Configuration 6] The system of any of configurations 1-5, wherein the one or more input parameters include one or more parameters defining a position of solidified or unsolidified material in the detection range of the sensor system. [Configuration 7] 7. The system of any of configurations 1-6, wherein the one or more input parameters include one or more of thermal conductivity, radiation properties, and temperature of the unsolidified material and the solidified material in the detection range of the sensor system, respectively. [Configuration 8] 8. The system of any of configurations 1 to 7, wherein the system is configured to calculate the one or more input parameters as a function of time. [Configuration 9] The system of any of configurations 1 to 8, wherein the system is configured to calculate the one or more input parameters for the specified material layer based on one or more output parameters output based on the simulation for one or more material layers underlying the specified material layer. [Configuration 10] 10. The system of claim 9, wherein the system is configured to calculate the one or more input parameters for the first plurality of material layers based on one or more output parameters output based on the simulation for a second plurality of material layers underlying the first plurality of material layers. [Configuration 11] The system of configuration 9 or 10, wherein the system is configured to calculate the one or more input parameters for the material layer or layers based on one or more parameters indicative of when one or more underlying layers have been or are being irradiated. [Configuration 12] 12. The system of embodiment 11, wherein the one or more parameters indicative of when the one or more underlying layers have been or are being illuminated are position dependent parameters. [Configuration 13] 13. The system of any of configurations 1-12, wherein the one or more input parameters include one or more of a radiation beam power, a scanning speed of a radiation beam used in the additive layer manufacturing process, a hatch distance used in the additive layer manufacturing process, and a material layer thickness of the material. [Configuration 14] The one or more input parameters are: the diameter of the radiation beam, heat dissipation during the additive layer manufacturing process; the physical properties of the material to be solidified; - the optical surface and / or the scanning device and / or the laser device and / or the layer deposition device of the additive layer production apparatus, and / or the gas flow and / or the temperature distribution in the additive layer production apparatus, and - a scanning strategy used within said additive layer manufacturing process, in particular the position of said radiation beam on the material layer and / or the deflection angle of said radiation beam and / or the vector angle and / or vector length of a vector for scanning a radiation beam across said material layer and / or a scanning direction for scanning said radiation beam across a material layer relative to the direction of gas flow, 14. A system according to any one of configurations 1 to 13, relating to or including one or more of the above. [Configuration 15] the processing unit includes an artificial neural network; The generation of the model comprises: generating a dataset comprising the sensor data and the one or more input parameters; splitting the data set into a test data set and a training data set; selecting a neural network model for said artificial neural network; training the artificial neural network based on the neural network model using the training data set; and evaluating the neural network model using the test data set. [Configuration 16] The evaluation may include: specifying requirements for said artificial neural network; and if a metric associated with the artificial neural network does not meet the requirement, modifying a structure and / or topology of the artificial neural network until the requirement is met. [Configuration 17] 17. The system of claim 16, wherein the generating of the model includes training the artificial neural network based on the altered structure and / or topology of the artificial neural network. [Configuration 18] 18. The system of any one of configurations 1 to 17, wherein the system is configured to simulate sensor data output by the sensor system based on the model and the one or more input parameters. [Configuration 19] 20. The system of claim 18, wherein the sensor data output by the sensor system includes the sensor data used by the processing device to generate the model. [Configuration 20] 20. The system of any one of configurations 18 to 19, wherein the sensor data output by the sensor system includes sensor data output by the sensor system after the model has been generated. [Configuration 21] The system comprises: comparing the sensor data output by the sensor system with the simulated sensor data; 21. The system of claim 18, 19 or 20, configured to detect anomalies and / or process errors in the additive manufacturing process based on the comparison. [Configuration 22] 22. The system of embodiment 21, wherein the system is configured to perform the comparison in-situ during the additive layer manufacturing process. [Structure 23] 23. The system of claim 22, wherein the system is configured to provide a signal to the additive manufacturing apparatus based on the comparison to adapt the current additive manufacturing process, in particular by stopping the current additive manufacturing process. [Structure 24] 24. The system of any of claims 1 to 23, wherein the system is configured to generate a database including the generated model for subsequent use in the simulation. [Configuration 25] 25. The system of any of configurations 1-24, further configured to adapt process parameters for the additive layered manufacturing process and / or a shape of the three-dimensional workpiece manufactured using the additive layered manufacturing process based on the simulation model. [Configuration 26] 16. The system of claim 15, wherein the training of the artificial neural network is based on minimizing a mean absolute error between the sensor data and simulation data associated with the simulation of the process monitoring. [Configuration 27] said training of said artificial neural network being based on a plurality of topologies of said artificial neural network; 16. The system of claim 15, wherein each of the topologies differs from the other topologies by a number of hidden layers and / or a number of neurons per hidden layer. [Configuration 28] 28. The system of claim 27, wherein the system is configured to train the topology of the artificial neural network until a minimum mean absolute error is reached. [Configuration 29] 16. The system of claim 15, wherein the artificial neural network comprises a multidimensional artificial neural network. [Configuration 30] 30. A computer program product comprising: a program code portion for simulating process monitoring by a sensor system in an additive manufacturing process using the model generated using the system of any one of configurations 1 to 29. [Structure 31] 31. The computer program product of configuration 30 stored on a computer-readable storage medium. [Structure 32] 1. A method for determining parameters influencing an additive manufacturing process used to solidify a material from which a three-dimensional workpiece is manufactured, comprising: determining an area and / or volume of one or more layers of material; determining, for a particular point in time, what portions of the area and / or volume of the one or more layers of material have already been solidified; determining the parameters influencing an additive manufacturing process as a function of the portions of the area and / or volume of the one or more layers of material that are determined to be already solidified. [Structure 33] 33. The method of claim 32, wherein the parameter is determined based on a first ratio between solidified and unsolidified material in the region and / or in the volume for the particular time point. [Structure 34] 34. The method of claim 32 or 33, wherein the parameters are determined based on a second ratio between solidified material and to-be-solidified material in the region and / or in the volume for the particular time point to produce the three-dimensional workpiece. [Structure 35] 35. The method of any of claims 32-34, wherein the parameters are determined based on a location of previously solidified material in the region and / or in the volume. [Structure 36] 36. The method of any of aspects 32-35, wherein the area and / or volume is greater than the area and / or volume, respectively, of the three-dimensional workpiece fabricated using the additive layer manufacturing process. [Structure 37] 1. A method for simulating process monitoring, in particular melt pool monitoring, by a sensor system, in particular a melt pool monitoring system, in an additive layer manufacturing process used for solidifying a material from which a three-dimensional workpiece is manufactured, comprising: determining a parameter using the method according to any one of aspects 32 to 36; providing the parameters to a model for simulating the process monitoring. [Structure 38] determining reference parameters using the determined parameters and input parameters; The input parameters are: parameters relating to the environment of the apparatus used for said additive manufacturing process; parameters related to the physical properties of the material to be solidified; parameters related to said process monitoring; parameters relating to the apparatus used for the additive manufacturing process; and parameters associated with said additive manufacturing process; one or more of 38. The method of claim 37, further comprising using the reference parameters to interpret a sensor signal output by the sensor system. [Structure 39] 39. The method of claim 38, wherein the reference parameters are determined using an artificial neural network. [Configuration 40] 40. The method of claim 39, wherein inputs to the artificial neural network include one or more constants and / or one or more one-dimensional time series. [Structure 41] The sensor signal is (i) to determine parameters influencing said additive manufacturing process; and / or (ii) as the input parameters to the artificial neural network 41. The method according to claim 39 or 40, wherein [Structure 42] A computer program product comprising program code portions for performing the method of any of configurations 32-41 when executed on one or more computing devices. [Structure 43] 43. The computer program product of configuration 42 stored on a computer-readable storage medium. [Configuration 44] 1. An apparatus for manufacturing a three-dimensional workpiece using additive layer build techniques, comprising: a carrier configured to receive material for manufacturing the three-dimensional workpiece; a material supply device configured to supply material to the carrier and / or a preceding layer of material on the carrier; a layer deposition mechanism for forming the dispensed material into a material layer on the carrier and / or the preceding material layer on the carrier; a solidification device configured to solidify the material provided on the carrier and / or the preceding material layer on the carrier to produce the three-dimensional workpiece; a gas supply configured to supply a shielding gas to an area of the layer of material being solidified by the solidification device; a process chamber containing the gas supply and the solidification equipment; a monitoring system for monitoring the area where the material is solidifying; An apparatus comprising: a system according to any one of configurations 1 to 29. [Configuration 45] The apparatus of configuration 44, further comprising the computer program product of configurations 30, 31, 42 or 43. [Configuration 46] providing sensor data relating to the process being monitored or previously monitored by a sensor system during an additive manufacturing process; Providing a model for simulating monitoring of the process, the model comprising: (i) generated using the system according to any one of configurations 1 to 29 or the apparatus according to configurations 44 or 45; and / or (ii) provided through a computer program product according to structure 30, 31, 42 or 43; and / or (iii) Generated using parameters determined according to any of configurations 32 to 36, particularly in combination with any of configurations 37 to 41; and The method further comprising: analyzing the sensor data based on a comparison between the sensor data and simulated sensor data using the model. [Configuration 47] The sensor data relating to the process monitored by the sensor system, Melt pool characteristics, radiation emanating from the melt pool, in particular thermal radiation emanating from the melt pool; radiation arising from solidified materials, a characteristic of a plasma generated from the additive manufacturing process used to generate the melt pool, the plasma resulting from material evaporated from the melt pool; radiation originating from the plasma, in particular thermal radiation originating from the plasma, and radiation reflected and / or diffracted from the melt pool and / or the plasma, in particular laser radiation originating from a laser device used to generate the melt pool; The method of claim 46, further comprising sensor data relating to one or more of the following: [Configuration 48] 48. A method comprising providing sensor data analyzed using the method of any one of claims 46 to 47. [Configuration 49] A computer-readable storage medium storing sensor data analyzed according to the method of any one of claims 46 to 47.
Claims
1. A system for simulating process monitoring by a sensor system in an additive layer manufacturing process, in particular a selective laser melting process, used to solidify a material from which a three-dimensional workpiece is manufactured, comprising: a data input device coupleable to a data output device of the sensor system and configured to receive sensor data from the sensor system relating to the process monitored by the sensor system, the data input device comprising: (i) associated with said additive manufacturing process; and / or (ii) relating to an additive manufacturing apparatus used for said additive manufacturing process; and / or (iii) associated with the process monitoring and / or the sensor system; a data input device further configured to receive one or more input parameters; a processing device coupled to the data input device and configured to generate a model from the sensor data and the one or more input parameters for simulating the process monitoring.
2. the process monitoring includes monitoring a melt pool; the sensor system includes a melt pool monitoring system; the model for simulating the monitoring of the process includes a model for simulating the monitoring of the melt pool; The sensor data related to the melt pool monitored by the melt pool monitoring system includes: characteristics of the melt pool; radiation emanating from the melt pool, in particular thermal radiation emanating from the melt pool; radiation arising from solidified materials, a characteristic of a plasma generated from the additive manufacturing process used to generate the melt pool, the plasma resulting from material evaporated from the melt pool; radiation originating from the plasma, in particular thermal radiation originating from the plasma, and radiation reflected and / or diffracted from the melt pool and / or the plasma, in particular laser radiation originating from a laser device used to generate the melt pool; The system of claim 1 , further comprising sensor data relating to one or more of:
3. 3. The system of claim 1 or 2, wherein the one or more input parameters are related to or calculated based on a distribution and / or arrangement of areas and / or volumes of unsolidified and solidified material relative to one another in a detection range of the sensor system.
4. The system of any one of claims 1 to 3, wherein the one or more input parameters are related to or calculated based on a ratio between unsolidified and solidified material in the detection range of the sensor system.
5. 5. The system of claim 4, wherein the ratio between unsolidified and solidified material in the detection area comprises a ratio between unsolidified and solidified material in the detection area over a predetermined number of material layers.
6. The system of any one of claims 1 to 5, wherein the one or more input parameters include one or more parameters defining a position of solidified or unsolidified material in the detection range of the sensor system.
7. 7. The system of claim 1, wherein the one or more input parameters comprise one or more of thermal conductivity, radiation properties and temperature of the unsolidified and solidified materials, respectively, in the detection range of the sensor system.
8. The system of any one of claims 1 to 7, wherein the system is configured to calculate the one or more input parameters as a function of time.
9. The system of any one of claims 1 to 8, wherein the system is configured to calculate the one or more input parameters for the defined material layer based on one or more output parameters output based on a simulation for one or more material layers underlying the defined material layer.
10. 10. The system of claim 9, wherein the system is configured to calculate the one or more input parameters for the first plurality of material layers based on one or more output parameters output based on the simulation for a second plurality of material layers underlying the first plurality of material layers.
11. 11. The system of claim 9 or 10, wherein the system is configured to calculate the one or more input parameters for the material layer or layers based on one or more parameters indicative of when one or more underlying layers have been or are being irradiated.
12. The system of claim 11 , wherein the one or more parameters indicative of when the one or more underlying layers have been or are being illuminated are position dependent parameters.
13. The system of any one of claims 1 to 12, wherein the one or more input parameters include one or more of a radiation beam power, a scanning speed of a radiation beam used in the additive layer manufacturing process, a hatch distance used in the additive layer manufacturing process, and a material layer thickness of the material.
14. The one or more input parameters are: the diameter of the radiation beam, heat dissipation during the additive layer manufacturing process; the physical properties of the material to be solidified; - the optical surface and / or the scanning device and / or the laser device and / or the layer deposition device of the additive layer production apparatus, and / or the gas flow and / or the temperature distribution in the additive layer production apparatus, and - a scanning strategy used within the additive layer manufacturing process, in particular the position of the radiation beam on the material layer and / or the deflection angle of the radiation beam and / or the vector angle and / or the vector length of a vector for scanning the radiation beam across the material layer and / or a scanning direction for scanning the radiation beam across a material layer relative to the gas flow direction, A system according to any one of claims 1 to 13, relating to or including one or more of:
15. the processing unit includes an artificial neural network; The generation of the model comprises: generating a dataset comprising the sensor data and the one or more input parameters; splitting the data set into a test data set and a training data set; selecting a neural network model for said artificial neural network; training the artificial neural network based on the neural network model using the training data set; and evaluating the neural network model using the test data set.
16. The evaluation may include: specifying requirements for said artificial neural network; and if a metric associated with the artificial neural network does not meet the requirement, modifying a structure and / or topology of the artificial neural network until the requirement is met.
17. 17. The system of claim 16, wherein the generating of the model comprises training the artificial neural network based on the altered structure and / or topology of the artificial neural network.
18. The system of any one of claims 1 to 17, wherein the system is configured to simulate sensor data output by the sensor system based on the model and the one or more input parameters.
19. The system of claim 18 , wherein the sensor data output by the sensor system includes the sensor data used by the processing unit to generate the model.
20. 20. The system of claim 18 or 19, wherein the sensor data output by the sensor system comprises sensor data output by the sensor system after the model has been generated.
21. The system comprises: comparing the sensor data output by the sensor system with the simulated sensor data; 21. The system of claim 18, 19 or 20, configured to detect anomalies and / or process errors in the additive layer manufacturing process based on the comparison.
22. 22. The system of claim 21, wherein the system is configured to perform the comparison in-situ during the additive layer manufacturing process.
23. 23. The system of claim 22, wherein the system is configured to provide a signal to the additive manufacturing apparatus based on the comparison to adapt the current additive manufacturing process, in particular by stopping a current additive manufacturing process.
24. A system according to any preceding claim, wherein the system is configured to generate a database including the generated model, for subsequent use in a simulation.
25. 25. The system of claim 1, further configured to adapt process parameters for the additive layer manufacturing process and / or a shape of the three-dimensional workpiece manufactured using the additive layer manufacturing process based on the model for simulating the process monitoring.
26. 16. The system of claim 15, wherein the training of the artificial neural network is based on minimizing a mean absolute error between the sensor data and simulation data associated with a simulation of the process monitoring.
27. said training of said artificial neural network being based on a plurality of topologies of said artificial neural network; 16. The system of claim 15, wherein each of the plurality of topologies differs from other topologies of the plurality of topologies by a number of hidden layers and / or a number of neurons per hidden layer.
28. The method of claim 27, wherein the training of the artificial neural network is based on minimizing a mean absolute error between the sensor data and simulation data associated with the simulation of the process monitoring; 28. The system of claim 27, wherein the system is configured to train the multiple topologies of the artificial neural network until a minimum mean absolute error is reached.
29. 16. The system of claim 15, wherein the artificial neural network comprises a multi-dimensional artificial neural network.
30. A computer program product comprising program code portions for simulating process monitoring by a sensor system in an additive layer manufacturing process using the model generated using the system according to any one of claims 1 to 29.
31. 31. The computer program product of claim 30 stored on a computer readable storage medium.
32. 1. A method for determining parameters influencing an additive layer manufacturing process used to solidify a material from which a three-dimensional workpiece is manufactured, comprising: determining an area and / or volume of one or more layers of material; determining, for a particular point in time, what portions of the area and / or volume of the one or more layers of material have already been solidified; determining the parameters influencing an additive layer manufacturing process as a function of the portions of the area and / or volume of the one or more layers of material that are determined to be already solidified.
33. 33. The method of claim 32, wherein the parameter is determined based on a first ratio between solidified and unsolidified material in the region and / or in the volume for the particular point in time.
34. 34. The method of claim 32 or 33, wherein the parameter is determined based on a second ratio between solidified material in the region and / or in the volume for the particular time point to produce the three-dimensional workpiece.
35. A method according to any one of claims 32 to 34, wherein the parameters are determined based on the location of previously solidified material in the region and / or in the volume.
36. The method of any one of claims 32 to 35, wherein the area and / or volume is larger than the area and / or volume, respectively, of the three-dimensional workpiece manufactured using the additive layer manufacturing process.
37. 1. A method for simulating process monitoring, in particular melt pool monitoring, by a sensor system, in particular a melt pool monitoring system, in an additive layer manufacturing process used to solidify a material from which a three-dimensional workpiece is manufactured, comprising: Determining a parameter using the method according to any one of claims 32 to 36; providing the parameters to a model for simulating the process monitoring.
38. determining reference parameters using the determined parameters and input parameters; The input parameters are: parameters relating to the environment of the apparatus used for said additive manufacturing process; parameters related to the physical properties of the material to be solidified; parameters related to said process monitoring; parameters relating to the apparatus used for the additive manufacturing process; and parameters associated with said additive manufacturing process; and 38. The method of claim 37, further comprising using the reference parameters to interpret a sensor signal output by the sensor system.
39. 39. The method of claim 38, wherein the reference parameters are determined using an artificial neural network.
40. 40. The method of claim 39, wherein inputs to the artificial neural network include one or more constants and / or one or more one-dimensional time series.
41. The sensor signal is (i) to determine parameters influencing said additive manufacturing process; and / or (ii) as the input parameters to the artificial neural network 41. The method according to claim 39 or 40, wherein
42. A computer program product comprising program code portions for performing the method according to any one of claims 32 to 41 when said computer program product is executed on one or more computing devices.
43. 43. The computer program product of claim 42 stored on a computer readable storage medium.
44. 1. An apparatus for manufacturing a three-dimensional workpiece using additive layer build techniques, comprising: a carrier configured to receive material for manufacturing the three-dimensional workpiece; a material supply device configured to supply material to the carrier and / or to a preceding layer of material on the carrier; a layer deposition mechanism for forming the dispensed material into a material layer on the carrier and / or the preceding material layer on the carrier; a solidification device configured to solidify the material provided on the carrier and / or the preceding material layer on the carrier to produce the three-dimensional workpiece; a gas supply configured to supply a shielding gas to an area of the layer of material being solidified by the solidification device; a process chamber containing the gas supply and the solidification equipment; a monitoring system for monitoring the area where the material is solidifying; and a system according to any one of claims 1 to 29.
45. 45. The apparatus of claim 44, further comprising a computer program product comprising program code portions for simulating process monitoring by a sensor system in an additive layer manufacturing process using the model generated using the system, the computer program product stored on a computer readable medium, or the computer program product of claim 42 or 43.
46. providing sensor data relating to the process being monitored or previously monitored by a sensor system during an additive manufacturing process; Providing a model for simulating monitoring of the process, the model comprising: (i) generated using the system according to any one of claims 1 to 29 or the apparatus according to claim 44 or 45; and / or (ii) and / or provided through a computer program product according to claim 30, 31, 42 or 43; (iii) Generated using parameters determined according to any one of claims 32 to 36, in particular in combination with any one of claims 37 to 41. and The method further comprising: analyzing the sensor data based on a comparison between the sensor data and simulated sensor data using the model.
47. The sensor data relating to the process monitored by the sensor system, Melt pool characteristics, radiation emanating from the melt pool, in particular thermal radiation emanating from the melt pool; radiation arising from solidified materials, a characteristic of a plasma generated from the additive manufacturing process used to generate the melt pool, the plasma resulting from material evaporated from the melt pool; radiation originating from the plasma, in particular thermal radiation originating from the plasma, and radiation reflected and / or diffracted from the melt pool and / or the plasma, in particular laser radiation originating from a laser device used to generate the melt pool; 47. The method of claim 46, further comprising sensor data relating to one or more of:
48. 48. A method comprising providing sensor data analysed using the method of claim 46 or 47.
49. 48. A computer readable storage medium storing sensor data analysed according to the method of claim 46 or 47.
Citation Information
Patent Citations
Real-time adaptive control of additive manufacturing processes using machine learning
JP2020527475A
Visualization of additive manufacturing process data
US20160179064A1
Real-time adaptive control of additive manufacturing processes using machine learning
WO2018217903A1
