Method and system for predicting build material layer overheating in additive manufacturing system

Abstract

A method and system for predicting build material layer overheating in an additive manufacturing system is provided. The method includes accessing a representation of a three-dimensional object as a plurality of cross-sectional layers for printing in an additive manufacturing system; accessing scan path data of a region of a selected one of the plurality of cross-sectional layers; subdividing the region into a plurality of units; and for each unit, generating a feature vector based on the representation of the object and the scan path data, and evaluating the trained prediction model based on the feature vector to obtain an output comprising a predicted melt time of the unit.

Description

Technical Field

[0001] This disclosure relates to a method and system for predicting the melting time of building materials during the manufacture of an object in an additive manufacturing system. Background Technology

[0002] Additive manufacturing is a technology that builds three-dimensional objects based on digital models. Compared to conventional subtractive manufacturing methods such as milling or machining, which create the final product by removing material, additive manufacturing adds material layer by layer. The building materials can include plastics, metals, ceramics, or many other types of materials.

[0003] Powder bed fusion (PBF) technology is an additive manufacturing technique that facilitates the production of high-quality parts with complex geometries and can be optimized for specific properties such as weight and strength. A PBF system includes a build platform onto which powdered build material is deposited layer by layer. Once a layer of build material has been deposited, a high-energy source, such as a laser beam, is controlled to selectively heat the powder material, causing it to heat and solidify to form a cross-sectional layer of the object.

[0004] Common PBF (Pulse-on-Pack) methods include selective laser sintering (SLS) and selective laser melting (SLM). In SLS, a laser systematically sinters the particles of a polymer-based powder and builds the part layer by layer. SLM systems can manipulate metal alloy powders. SLM is used in the production of metal parts in the aerospace, automotive, and healthcare industries.

[0005] In a PBF system, the high-energy source is software-controlled, guiding it to heat the build-up material along a scanning path. The scanning path is a predetermined path followed by the heat source to create layers of the object. The scanning path taken by the heat source, as well as its traversal time and power, plays a crucial role in determining the material properties of the build-up and avoiding manufacturing defects.

[0006] Creating the optimal scan path is a complex task. Conventional scan path algorithms use a globally optimized set of printing parameters derived from testing a large number of part samples. These parameters are chosen to globally optimize material properties such as density and tensile strength, as well as the build time of the entire build. Such globally optimized parameters may lead to specific printing problems in localized parts. Parameters obtained through trial and error cannot guarantee local build quality for geometries with complex topologies and fine internal structures. Build parameters that are optimal for printing simple test parts may be suboptimal, or even problematic, for specific geometric features in printing actual parts.

[0007] A major cause of localized problems in components is localized overheating. Overheating can occur, for example, when an energy source is repeatedly directed to the same area, or when an energy source is directed to a point for an excessively long period. Localized overheating has also been observed in overhanging areas. Overhanging areas are more prone to overheating because they lack the support of an underlying layer, as there is no building material beneath the overhanging layer to conduct heat away, leading to heat buildup.

[0008] Finite element analysis (FEA) simulations can be used to analyze and optimize various aspects of the PBF (Physical Building Fabric) process. FEA can simulate the thermal behavior of the PBF process, including predicting temperature gradients. This allows engineers to identify potential overheated areas in the build. Nevertheless, the application of FEA remains limited. Performing FEA at the path scale is inefficient: due to the small layer thicknesses and scanline spacing involved in the build, the data size of a single build file can reach terabytes. Therefore, using conventional FEA to detect localized problems in complex part geometries can take months, making it impractical. Summary of the Invention

[0009] One objective of this disclosure is to enable the rapid identification of overheated areas when manufacturing objects in an additive manufacturing system.

[0010] The foregoing and other objectives are achieved through the features of the independent claims. Further implementations are apparent from the dependent claims, the specification, and the drawings.

[0011] According to a first aspect, a computer-implemented method is provided for predicting overheating of a build material layer in an additive manufacturing system. The method includes: accessing a three-dimensional object as a representation of a plurality of cross-sectional layers for printing in the additive manufacturing system; accessing scan path data of a region of a selected layer among the plurality of cross-sectional layers; and subdividing the region into a plurality of cells. Furthermore, the method includes, for each cell: generating a feature vector based on the representation of the object and the scan path data; and evaluating a trained prediction model based on the feature vector to obtain an output including a predicted melting time for the cell.

[0012] According to a second aspect, a computer implementation method is provided for training a prediction model for predicting the melting time of build material in deposited build material layers in an additive manufacturing system. The method includes: accessing a set of model parameters of the prediction model; accessing object data and print data of the object printed in the additive manufacturing system, the print data including a representation of the object as a plurality of cross-sectional layers and scan path data for each of the plurality of cross-sectional layers; accessing training data of the object, the training data including simulations of the melting time for each of the plurality of cross-sectional layers; subdividing the object into a plurality of units based on the object data; generating a feature vector for each of the plurality of units based on the print data; evaluating the prediction model to obtain a predicted melting time for each unit; and modifying the set of model parameters based on the training data and the predicted melting time in each unit.

[0013] According to a third aspect, a data processing system is provided, including a processor and a memory. The memory stores instructions that, when executed by the processor, cause the processor to perform the method according to the first aspect.

[0014] According to a fourth aspect, a non-transitory computer-readable storage medium is provided. The non-transitory computer-readable storage medium contains program code that, when executed by a processor, provides instructions to perform the method according to the first aspect.

[0015] In a first implementation of the method according to the first aspect, the melting time includes the maximum time period during which the building material of the unit exceeds the melting point of the building material.

[0016] In a second implementation, the method includes: comparing the predicted melting time with a threshold; and identifying the region as an overheated region when the melting time exceeds the threshold.

[0017] In a third implementation, the method includes: modifying the scan path in response to identifying the region as an overheated region.

[0018] In the fourth implementation, modifying the scan path includes modifying the power source, scan speed, delay time, or regenerating the scan path.

[0019] In the fifth implementation, the trained prediction model includes a neural network model.

[0020] In the sixth implementation, the feature vector includes the energy density value of the cell.

[0021] In the seventh implementation, the feature vector includes values ​​determined based on the energy source timing data of adjacent units in the cross-sectional layer.

[0022] In the eighth implementation, the selected cross-sectional layer is chosen based on the output of the thermal analysis test applied to each of the plurality of cross-sectional layers.

[0023] These and other aspects of this disclosure will become apparent in the embodiments described below. Attached Figure Description

[0024] To gain a more complete understanding of this disclosure and its advantages, reference is now made to the following description taken in conjunction with the accompanying drawings, wherein:

[0025] Figure 1 It is a schematic diagram based on an example computing system;

[0026] Figure 2 It is based on an example, a schematic perspective view showing the scan path on an object;

[0027] Figure 3A It is based on an example, schematic perspective view showing the part to be manufactured;

[0028] Figure 3B It is based on an example Figure 3A A cross-sectional slice of the part shown;

[0029] Figure 3C It is based on an example Figure 3B The scanning strategy for the cross-sectional slices shown;

[0030] Figure 3D It is based on an example Figure 3B The overheating condition on the cross-sectional slice shown;

[0031] Figure 4 is a schematic diagram of a machine learning unit (MLU) based on an example;

[0032] Figure 5 This is a flowchart illustrating an example method for training a prediction model to predict melting time.

[0033] Figure 6 An example of a training object used to train a prediction model is shown;

[0034] Figure 7 It is a flowchart based on an example method for predicting the melting time of building materials;

[0035] Figure 8 This is a simplified schematic diagram based on an example computing system. Detailed Implementation

[0036] The exemplary embodiments described below are detailed enough to enable those skilled in the art to emulate and implement the systems and processes described herein. It is important to understand that embodiments may be provided in many alternative forms and should not be construed as limited to the examples set forth herein.

[0037] Therefore, while embodiments may be modified in various ways and present various alternative forms, specific embodiments thereof are shown in the accompanying drawings and described in detail below as examples. There is no intention to limit oneself to the specific forms disclosed. Rather, all modifications, equivalents, and alternatives falling within the scope of the appended claims should be included. Elements of the exemplary embodiments are consistently denoted by the same reference numerals throughout the drawings and detailed description (where applicable).

[0038] The terminology used herein to describe embodiments is not intended to be limiting. The articles “a,” “an,” and “the” are singular when they refer to a single object, but their use in the singular form herein should not preclude the existence of multiple objects of reference. In other words, unless the context clearly indicates otherwise, an element referred to in the singular form may be one or more. It will be further understood that the terms “comprising,” “including,” and / or “containing,” as used herein, specify the presence of the stated feature, item, step, operation, element, and / or component, but do not preclude the presence or addition of one or more other features, items, steps, operations, elements, components, and / or combinations thereof.

[0039] Unless otherwise defined, all terms used herein (including technical and scientific terms) should be interpreted as having their conventional meaning in the art. It will be further understood that commonly used terms should also be interpreted as having their conventional meaning in the relevant technical field, rather than idealized or overly formal meanings, unless explicitly defined herein.

[0040] The methods and systems described in this paper utilize machine learning to rapidly identify localized overheated areas in additive manufacturing. Furthermore, the methods described in this paper provide intelligent scan path optimization when needed to prevent problems in specific layers and regions.

[0041] According to the example, each layer is subdivided into a mesh structure containing multiple grid cells, where the size of the cells is determined by the scanline spacing of the energy source. A machine learning simulation is applied to a cell, called a Machine Learning Unit (MLU), which includes the cell to which the energy source is applied and its neighboring cells.

[0042] The method described in this paper efficiently simulates the thermal behavior of scan paths to identify overheated regions. The simulation can be executed immediately after the scan path generation algorithm outputs the scan path. The simulation can be executed in parallel with the generation of scan paths for subsequent layers to minimize the simulation's overhead on print job generation time. The machine learning unit (MLU)-based simulation described in this paper can be deployed on any material and multiple laser scan paths with any combination of scan path strategies.

[0043] Scan path optimization strategies are used to simulate and predict problematic areas. Specifically, in identified overheated areas, the scan strategy or construction parameters can be modified. This might include, for example, changing laser power, scan speed, or delay time, locally reordering scan vectors, or even regenerating the scan path.

[0044] Figure 1 This is a simplified schematic diagram of an example computing system 100. The computing system 100 can be used to generate and optimize scan paths in additive manufacturing. In the example described herein, the computing system 100 may be an application server, computing node, desktop computer, laptop computer, smartphone or other portable device, embedded device, or similar device. In other cases, the computing system 100 may represent multiple computing systems in a distributed or networked configuration.

[0045] The computing system 100 can be accessed via wired and wireless interfaces ( Figure 1 (Not shown) The computing system 100 communicates with other systems. Examples of devices and systems that can communicate with the computing system 100 via wired or wireless connections include keyboards, mice, monitors, touchscreen interfaces, trackballs, cameras, microphones, scanners, printers, speakers, pointers, touchpads, drawing tablets, joysticks, and motion sensing devices that capture motion gestures. Some devices function as user input devices, facilitating user input into the computing system 100, while others function as output devices.

[0046] The computing system 100 can be configured to implement computer-aided design (CAD) and computer-aided manufacturing (CAM) applications for manufacturing objects in an additive manufacturing system. It should be understood that the CAD and CAM applications can be used as graphical user interfaces to facilitate interaction with computer-generated object models.

[0047] According to an example, the computing system 100 includes a storage device 110. The storage device 110 is configured to store object data for an object to be manufactured in an additive manufacturing system. The storage device 110 is communicatively coupled to a build processor 120. The build processor 120 is configured to access the object data from the storage device 110 and generate print data for printing the object in the additive manufacturing system. The build processor 120 may be in the form of dedicated hardware or software implemented by a general-purpose processor. The processor may be in the form of a single-core or multi-core processor system. The print data includes subdividing the object into multiple cross-sectional layers and scan path data specifying scan paths for controlling the energy source of the additive manufacturing system during printing.

[0048] Figure 2 This is a schematic perspective view 200 showing the scanning path in an additive manufacturing system. Figure 2 An energy source 210, as part of an additive manufacturing system, is shown. The energy source 210 may be a laser or other form of high-energy source, such as an electron beam. The energy source 210 is controlled to direct energy to a layer 220 containing deposited build material. The layer 220 may be deposited on the build platform of the additive manufacturing system.

[0049] The additive manufacturing system controls the energy source 210 to move along the scanning path 230 indicated by the arrow in layer 220, thereby applying energy to layer 220 along the scanning path 230. The scanning path 230 is... Figure 1 The build processor 120 shown is generated by the build processor. Figure 2 In this context, the scan path 230 is a parallel vector spanning the length of the layer 220. In other examples, the scan path may follow a meandering, zigzag, spiral, checkerboard, contour, island, or raster pattern. The distance d between adjacent scan paths is referred to herein as the scan line spacing.

[0050] Figure 3A It is based on an example, a schematic perspective view 300 showing a part 310 manufactured in an additive manufacturing system. Figure 3A A cross-sectional slice 320 of part 310 is also shown. A view of the same cross-sectional slice 320 is also shown. Figure 3B As shown in the diagram. The part 310 can be represented as object data in a computer-aided design system. The object data of part 310 can be stored in storage device 110 and accessed by build processor 120. The object data can be in stereolithography file format (STL) or additive manufacturing file format (AMF).

[0051] To illustrate the overheating problem, Figure 3CThe scanning strategy for the cross-sectional layer 320 of part 310 shown in Figure 3 is illustrated. Figure 3A In the middle, the scanning path forms a checkerboard pattern on the cross-sectional layer 320. Figure 3D The illustration shows the application of a laser with a uniform power distribution. Figure 3D The thermal effect of the scanning strategy shown. Under uniform power conditions, region 330 will overheat because region 330 is overhanging relative to the build platform during the manufacturing of part 300.

[0052] Figure 4 is a schematic diagram of a machine learning unit (MLU) 400 according to an example. The MLU 400 comprises a grid representation consisting of multiple cells 410. In the example shown in Figure 4, the MLU comprises a 3x3x3 cell grid. In other examples, the MLU 400 may comprise a 4x4x4 or 5x5x5 grid. The MLU 400 is defined for each point along each scan path in a sub-region of a layer. The center cell 420 represents the point where an energy source is being applied. Each MLU is associated with parameters. These parameters include energy source settings: power at the center cell 420 of the grid, scan rate and time, and the density of material in each cell.

[0053] Figure 4B A top view of the MLU 410 is shown. The data value t in each grid cell is also shown. i,j This represents the energy source timing from the j-th nearest neighbor to the center unit 420. A feature vector can be defined for each MLU. ,as follows: in This includes the total mass of the i-th layer.

[0054] When the MLU is defined on a larger grid, similar features can be specified. In some examples, the feature vector includes additional features to specify the energy density at the central cell 420:

[0055] In this equation, P includes the power of the energy source. It is the power absorption ratio of the material, V is the scanning speed, and d is the scanning velocity. hatch It is the scan line spacing, and L layer This refers to the layer thickness. In some cases, the feature vector may also include a material identifier.

[0056] Figure 5A flowchart of a method 500 for training a predictive model to predict the melting time of build material in a deposited build material layer in an additive manufacturing system is shown. In this document, the term melting time refers to the maximum time period during which the build material of the cell exceeds the melting point of the build material. Method 500 can be used in conjunction with other methods and systems described herein, particularly the MLU 400 and associated feature vectors described above. Based on the examples described herein, the predictive model may include a neural network model.

[0057] In block 510, method 500 includes accessing a set of model parameters of the prediction model. According to an example, the set of model parameters may include weights of a neural network. In block 520, the method includes accessing object data of an object and print data of printing the object in an additive manufacturing system. The print data includes a representation of the object as a plurality of cross-sectional layers and scan path data for each of the plurality of cross-sectional layers. The scan path data specifies a scan path of a heat source in the additive manufacturing system to selectively heat deposited build material layers, causing the build material to melt and fuse to form cross-sectional layers.

[0058] In block 530, the method includes accessing training data for the object, the training data comprising simulations of the melting time for each of the plurality of cross-sectional layers. According to the example described herein, the simulations may be finite element analysis (FEA) simulations in each layer. The training data can be obtained by evaluating the simulations in each of the plurality of cross-sectional layers.

[0059] In block 540, the method includes subdividing the object into multiple units based on the object data. In an example, subdividing the object into multiple units includes subdividing the object into multiple MLUs similar to the previously described MLU 400.

[0060] In block 550, a feature vector is generated for each cell based on the printed data. In the example, the feature vector may have a vector for each MLU as described above. In block 560, method 500 includes evaluating the prediction model to obtain a predicted melting time for each unit. Evaluating the prediction model may include evaluating a neural network model and observing an output containing the predicted melting times. In block 570, the method includes modifying the set of model parameters based on the training data and the predicted melting times in each unit. According to an example, modifying the set of model parameters may include iteratively modifying the model parameters via gradient descent to minimize the loss function of the neural network model.

[0061] Figure 6Example object 600 is shown that can be used to train the prediction model. Object 610 is an example of a rectangular block. Objects 620, 630, and 640 include blocks with tilted sides having different angles. In the example, method 500 can be repeated for rectangular blocks similar to object 610 (rotation angles between 0 and 90 degrees) and objects similar to objects 620, 630, and 640 with tilted sides having different angles between 15 and 30 degrees. Furthermore, method 500 can be repeated for objects made of different materials and a range of energy source powers and velocities. According to the example, the energy source power can be between 0.5 and 1.5 times the power reference value P. ref The speed can vary between 0.5 and 1.5 times the speed reference value V. According to the example, the speed can be between 0.5 and 1.5 times the speed reference value V. ref The changes between them.

[0062] Figure 7 A flowchart of a method 700 for predicting the melt time of a deposited build material layer in an additive manufacturing system is shown. Method 700 can be used in conjunction with other methods and systems described herein, particularly the previously described MLU 400, relevant feature vectors, and training method 500. According to the examples described herein, the prediction model may include a neural network model.

[0063] In block 710, method 700 includes accessing a three-dimensional object as a representation of a plurality of cross-sectional layers for printing in an additive manufacturing system. In block 720, the method includes accessing scan path data of a region of a selected layer among the plurality of cross-sectional layers. The region may include the entire selected cross-sectional layer or a portion of the layer. The scan path data specifies a scan path of an energy source in the additive manufacturing system to selectively heat the deposited build material layer, causing the build material to melt and fuse to form the selected cross-sectional layer.

[0064] In block 730, method 700 includes subdividing the region into multiple cells. According to an example, the region can be subdivided by specifying a grid, the grid size being equal to the scan line spacing of the energy source, as previously described.

[0065] In block 740, method 700 includes, for each cell, generating a feature vector based on the representation of the object and scan path data. The feature vector may be a previously defined vector. In block 750, the method includes evaluating a trained prediction model to obtain an output including the predicted melting time of the unit. The trained prediction model may be as described above in conjunction with method 500 and based on... Figure 6 A predictive model trained on the described objects and data.

[0066] Based on the example, the region can be determined to be an overheated region based on the evaluation of the prediction model. In this case, the scan path can be optimized, for example, by changing the power of the energy source, the scan speed, the delay time, or by regenerating a portion of the scan path in the region.

[0067] In one example, method 700 is repeated for each cross-sectional layer. However, in another example, a rapid approximate thermal analysis check can be performed on each layer before executing method 700, allowing the user to quickly identify which layers may be problematic and which are not.

[0068] The method and system described in this paper offer numerous advantages. In particular, the method leverages the power of machine learning techniques and large amounts of simulation data to detect and mitigate local build problems. This can be performed in parallel during the scan path generation process. Suboptimal or problematic paths can be corrected during path generation. The described method eliminates the need for expensive monitoring hardware for feedback control and reduces the number of test print iterations required to achieve a good build process and high part quality. This saves significant time, materials, and money, and enables the printing of high-quality parts with minimal increase in print time.

[0069] Figure 8 An example of a data processing system that can implement embodiments of the present disclosure is shown. The data processing system 800 includes a processor 810 connected to a local system bus 820. The local system bus connects the processor to main memory 830 and a graphics display adapter 840, which can be connected to a display 850. The data processing system can communicate with other systems via a wireless user interface adapter connected to the local system bus 820 or via a wired network (e.g., a local area network). Additional memory 860 can also be connected via the local system bus 820.

[0070] A suitable adapter, such as a wireless user interface adapter 870, is used for other peripheral devices, such as a keyboard 880 and a mouse 890 or other pointing devices, to allow the user to provide input to the data processing system. Other peripheral devices may include one or more I / O controllers, such as a USB controller, a Bluetooth controller, and / or a dedicated audio controller (connected to a speaker and / or microphone). It should also be understood that various peripheral devices can be connected to the USB controller (via various USB ports), including input devices (e.g., keyboard, mouse, touchscreen, trackball, camera, microphone, scanner), output devices (e.g., printer, speaker), or any other type of device capable of providing input or receiving output from the data processing system.

[0071] Furthermore, it should be understood that many devices referred to as input devices or output devices can both provide input for communication with the data processing system and receive its output. It should also be understood that other peripheral hardware connected to the I / O controller can include any type of device, machine, or component configured to communicate with the data processing system.

[0072] The operating system in the data processing system enables the system's output to be displayed to the user, and allows the user to interact with the system. Examples of operating systems that can be used in the data processing system include Microsoft Windows™, Linux™, UNIX™, iOS™, and Android™.

[0073] Furthermore, it should be understood that the data processing system 800 can be implemented in a network environment, a distributed system environment, a virtual machine in a virtual machine architecture, and / or a cloud environment. For example, the processor and related components may correspond to a virtual machine running in a virtual machine environment on one or more servers. Examples of virtual machine architectures include VMware ESCi, Microsoft Hyper-V, Xen, and KVM.

[0074] Those skilled in the art will understand that the hardware description of the data processing system 800 may vary depending on the specific implementation. For example, the data processing system 800 in this example may correspond to a computer, workstation, and / or server. However, it should be understood that alternative embodiments of the data processing system may be configured with corresponding or alternative components, such as in the form of a mobile phone, tablet computer, controller board, or any other system capable of processing data and implementing the functions and features described herein related to the operation of data processing systems, computers, processors, and / or controllers. The examples depicted are provided for illustrative purposes only and are not intended to imply any architectural limitations on this disclosure.

[0075] The data processing system 800 may be connected to a network (not part of the data processing system 800), which may be any public or private data processing system network or combination of networks known to those skilled in the art, including the Internet. The data processing system 800 may communicate with one or more other data processing systems (e.g., servers, which are not part of the data processing system 800) via the network. However, alternative data processing systems may correspond to multiple data processing systems implemented as part of a distributed system, where processors associated with multiple data processing systems may communicate via one or more network connections and may collectively perform tasks described as being performed by a single data processing system. Therefore, it should be understood that, when referring to a data processing system, such a system may be implemented by multiple data processing systems distributed across a network that communicate with each other.

[0076] The data processing system 800 is adapted to perform the methods according to the embodiments described herein. For example, a keyboard 880 and a mouse 890 may function as user input devices to receive information from a user, a processor 810 may be adapted to perform the steps of the method, and a display 850 may be adapted to display a specific view to the user. A computer product containing instructions may be provided that, when run on a computer such as the data processing system 800, causes the computer to perform the steps of the methods described in the above embodiments.

[0077] This disclosure is described with reference to flowcharts and / or block diagrams illustrating methods, apparatus, and systems. Although the flowcharts above show a specific execution order, the execution order may differ from what is depicted. A block associated with one flowchart may be combined with a block from another flowchart. In some examples, some blocks in the flowcharts may not be necessary, and / or additional blocks may be added.

[0078] This disclosure may be embodied in other specific apparatuses and / or methods. The described embodiments should be considered illustrative rather than restrictive in all respects. In particular, the scope of this disclosure is indicated by the appended claims rather than by the description and drawings herein. All modifications falling within the meaning and equivalents of the claims should be included within their scope.

Claims

1. A computer-implemented method for predicting overheating of a material layer in an additive manufacturing system, the method comprising: Access to a representation of a three-dimensional object, the representation being multiple cross-sectional layers for printing in the additive manufacturing system; Access scan path data of a region of a selected cross-sectional layer among the plurality of cross-sectional layers, the scan path data specifying the scan path of an energy source in the additive manufacturing system to selectively heat the deposited build material layer, causing the build material to melt and fuse to form the selected cross-sectional layer; The region is further subdivided into multiple units; Based on the representation of the 3D object and the scan path data, a feature vector is generated for each of the plurality of units; as well as Based on the feature vector, a trained prediction model is evaluated for each of the plurality of units to obtain an output including a predicted melting time for the corresponding unit.

2. The method of claim 1, wherein the predicted melting time includes the maximum time period during which the building material of the corresponding unit exceeds the melting point of the building material.

3. The method according to claim 1, further comprising: The predicted melting time is compared with a threshold. as well as When the predicted melting time exceeds the threshold, the region is identified as an overheated region.

4. The method according to claim 3, further comprising: In response to identifying the region as the overheated region, the scan path is modified.

5. The method of claim 4, wherein modifying the scan path includes modifying the power source, scan speed, delay time, or regenerating a portion of the scan path.

6. The method of claim 1, wherein the trained prediction model comprises a neural network model.

7. The method of claim 1, wherein the feature vector includes an energy density value for the corresponding unit.

8. The method of claim 7, wherein the feature vector comprises a value determined based on energy source timing data of adjacent units of the plurality of units in the cross-sectional layer.

9. The method of claim 1, wherein the selected cross-sectional layer is selected based on the output of a thermal analysis test applied to each of the plurality of cross-sectional layers.

10. A computer-implemented method for training a prediction model, said prediction model being used to predict the melting time of a build material layer deposited in an additive manufacturing system, the method comprising: Access the set of model parameters for the prediction model; Access to object data for an object and printing data for printing the object in the additive manufacturing system, wherein the printing data includes a representation of the object, the representation being a plurality of cross-sectional layers for printing in the additive manufacturing system, and the printing data includes the plurality of cross-sectional layers and scan path data for each of the plurality of cross-sectional layers, wherein the scan path data specifies a scan path of an energy source in the additive manufacturing system to selectively heat deposited build material layers, such that the build material melts and fuses to form the cross-sectional layers among the plurality of cross-sectional layers; Access training data for the object, wherein the training data includes simulations of the melting time for each of the plurality of cross-sectional layers; The object is subdivided into multiple units; Based on the printed data, a feature vector is generated for each of the plurality of units; The prediction model is evaluated to obtain the predicted melting time for each unit; as well as Based on the training data and the predicted melting time in each unit, the set of model parameters is modified.

11. The method of claim 10, further comprising: Repeat the steps according to claim 10 for multiple objects and energy source parameters.

12. The method of claim 10, wherein the prediction model is a neural network model, and wherein the model parameters include weights, and The simulation described therein is a finite element analysis (FEA) simulation.

13. A data processing system, comprising: A processor and a memory, the memory storing instructions that, when executed by the processor, cause the processor to perform the method according to any one of claims 1 to 12.

14. A non-transitory computer-readable storage medium comprising program code that, when executed by a processor, provides instructions to perform the method according to any one of claims 1 to 12.

Images