Distortion prediction for additive manufacturing using image analysis

Abstract

Examples described herein provide a method that includes performing image analysis on an image of a layer of an object manufactured by an additive manufacturing system to identify an exposed surface in the image of the layer. The method also includes performing a build simulation to generate a simulation distortion of the layer. The method also includes evaluating the build data to determine a value of an impact factor for the layer. The method further includes using a machine learning model to predict at least one of a predicted distortion or a predicted recoating disturbance for a next layer based at least in part on the image analysis, the building simulation, and the building data. The method further includes performing an action to alter manufacturing of the next layer based at least in part on at least one of the predicted distortion or the predicted recoater disturbance.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of U.S. Patent Application No. 17 / 727,362, filed on April 22, 2022, which is incorporated herein by reference in its entirety. Background Art

[0003] Embodiments described herein relate generally to additive manufacturing, and more particularly to techniques for distortion prediction for additive manufacturing using image analysis.

[0004] Additive manufacturing is a process that creates three-dimensional objects by depositing materials in successive layers. Various industries utilize additive manufacturing to create objects. These industries may include aerospace, automotive, consumer products, medical devices, oil and gas exploration and production, etc. Downhole exploration and production work involves deploying various sensors and tools into the earth to locate and extract hydrocarbons. Additive manufacturing can be used to form components of tools or entire tools used in downhole exploration and production work. Summary of the invention

[0005] Embodiments of the present invention relate to distortion prediction for additive manufacturing using image analysis.

[0006] A non-limiting example method includes performing image analysis on an image of a layer of an object manufactured by an additive manufacturing system to identify an exposed surface in the image of the layer. The method also includes performing a build simulation to generate a simulated distortion of the layer. The method also includes evaluating build data to determine a value of an impact factor for the layer. The method also includes using a machine learning model to predict at least one of a predicted distortion or a predicted recoater interference for a next layer based at least in part on the image analysis, the build simulation, and the build data. The method also includes implementing an action to change the manufacture of the next layer based at least in part on at least one of the predicted distortion or the predicted recoater interference.

[0007] A non-limiting example system includes a processing system including a memory and a processor for executing computer-readable instructions that control a processing device to perform operations. The operations include performing image analysis on an image of a layer of an object manufactured by an additive manufacturing system to identify an exposed surface in the image of the layer. The operations also include performing a build simulation to generate a simulated distortion of the layer. The operations also include comparing the exposed surface in the image of the layer with the simulated distortion of the layer. The operations also include using a machine learning model to predict the distortion of a next layer. The operations also include implementing actions to reduce distortion during manufacturing of the next layer based at least in part on the predicted distortion.

[0008] Other embodiments of the present invention implement the features of the above-described methods in computer systems and computer program products.

[0009] Additional technical features and benefits are achieved through the technology of the present invention. Embodiments and aspects of the present invention are described in detail herein and are considered part of the claimed subject matter. For a better understanding, reference is made to the specific embodiments and the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication in color drawing(s) will be provided by the Office upon request and payment of the necessary fee.

[0011] Referring now to the drawings, wherein like elements bear like reference numerals throughout the several views:

[0012] Figure 1 A block diagram of a system for distortion prediction for additive manufacturing using image analysis according to one or more embodiments described herein is shown;

[0013] Figure 2 A flow chart showing a method for distortion prediction for additive manufacturing using image analysis according to one or more embodiments described herein;

[0014] Figures 3A-3D An image showing a layer of an object manufactured by an additive manufacturing system according to one or more embodiments described herein;

[0015] Figure 3E showing an image of an exposed area according to one or more embodiments described herein;

[0016] Figure 3F An image showing streaks caused by recoater interference according to one or more embodiments described herein;

[0017] Figures 4A-4D shows an image of simulated distortion predicted by simulation according to one or more embodiments described herein;

[0018] Figure 5A and Figure 5B A schematic diagram showing a spreading simulation according to one or more embodiments described herein is shown;

[0019] Figure 6 A flow chart showing a method for distortion prediction for additive manufacturing using image analysis according to one or more embodiments described herein;

[0020] Figure 7 A flow chart showing a method for distortion prediction for additive manufacturing using image analysis according to one or more embodiments described herein;

[0021] Figure 8A flowchart of a method for training a machine learning model according to one or more embodiments described herein is shown; and

[0022] Fig. 9 A block diagram of a processing system for implementing the presently described techniques is shown, in accordance with one or more embodiments described herein. DETAILED DESCRIPTION

[0023] Additive manufacturing (sometimes referred to as "3D printing") includes a variety of manufacturing techniques (modalities) that allow an object to be "grown" layer by layer from a raw material (e.g., powder, wire, filament, etc.) into a desired shape, rather than removing material from a bar as in non-additive (i.e., subtractive) manufacturing methods. The main advantages of additive manufacturing are the relative simplicity of the manufacturing process (e.g., fewer steps, most of the process occurs within a single machine, there is no operator running the machine, etc.) and design freedom (i.e., the ability to form complex shapes).

[0024] During additive manufacturing, distortion (also referred to herein as "deformation") may occur. Distortion is the deflection of a part from its desired shape. Distortion leads to dimensional inaccuracies and even to recoater interference, which can cause quality issues and / or failed builds. Traditionally, distortion is not prevented during design by simulation alone.

[0025] Therefore, one or more embodiments for distortion prediction for additive manufacturing using image analysis are described herein. According to one or more embodiments described herein, a method for training a machine learning model (e.g., a recursive neural network) to predict how distortion affects the next layer of the build is provided. Specifically, once the machine learning model is trained, it can be used to perform real-time (or near real-time) distortion prediction using image analysis. By predicting distortion, additive manufacturing technology can be improved by reducing or eliminating distortion. According to one or more embodiments described herein, a method is provided that uses pre-established simulation results combined with real-time (or near real-time) image analysis in a trained machine learning model (e.g., a recursive neural network) to evaluate the actual distortion in the current layer and predict the distortion in the next layer. In some examples, the prediction can be used to infer recoating interference and / or dimensional accuracy. According to one or more embodiments described herein, process parameters of the additive manufacturing system can be controlled to reduce distortion, such as by reducing laser power on distorted areas, skipping layers in selected areas, etc., including combinations thereof.

[0026] Figure 1A block diagram of a processing system 100 for distortion prediction for additive manufacturing using image analysis according to one or more embodiments described herein is shown. The processing system 100 includes a processing device 102, a memory 104, a sensor 106, a data storage 108, a machine learning (ML) model training engine 110, an ML model inference engine 112, an image analysis engine 114, and / or a build simulation engine 116. Other configurations of the processing system 100 are possible, such that one or more of the components, engines, etc. may be removed and / or one or more additional components, engines, etc. may be added.

[0027] about Figure 1 The various components, engines, etc. described may be implemented as instructions stored on a computer-readable storage medium, hardware modules, dedicated hardware (e.g., dedicated hardware, application-specific integrated circuits (ASICs), application-specific special processors (ASSPs), field-programmable gate arrays (FPGAs), embedded controllers, hard-wired circuits, etc.), or one or more combinations of these. According to various aspects of the present disclosure, the engines described herein may be a combination of hardware and programming. The programming may be processor-executable instructions stored on a tangible memory, and the hardware may include a processing device 102 for executing those instructions. Thus, a system memory (e.g., memory 104) may store program instructions that, when executed by the processing device 102, implement the engines described herein. Other engines may also be utilized to include other features and functions described in other examples herein.

[0028] The processing device 102 may be any suitable processing device (or "processor") or a plurality of such devices. Examples of processing devices include a microprocessor, a microcontroller, a central processing unit (CPU), a graphics processing unit (GPU), a reduced instruction set computer (RISC) microprocessor, and the like, including combinations and / or multiples thereof. The processing device 102 may be coupled to a system memory (e.g., memory 104). Examples of system memory (such as memory 104) include a read-only memory (ROM), a random access memory (RAM), and the like, including combinations and / or multiples thereof. The sensor 106 may be any suitable device or combination of devices for collecting data. For example, the sensor 106 may be a camera, a temperature sensor, a vibration sensor, an optical sensor, and the like, including combinations and / or multiples thereof. The data storage 108 may be any suitable storage device or combination of devices for storing data (such as data collected by the sensor 106). For example, the data storage 108 may be a hard disk drive, a solid state disk, and the like, including combinations and / or multiples thereof.

[0029] The processing system 100 (using, for example, the processing device 102, the memory 104, the sensor 106 and / or the data storage 108) can use one or more of the ML model training engine 110, the ML model inference engine 112, the image analysis engine 114 and / or the construction simulation engine 116 to perform machine learning model training and inference, image analysis and / or construction simulation.

[0030] In addition, the cloud computing system can be in wired or wireless electronic communication with one or all of the elements of the processing system 100. Cloud computing can supplement, support, or replace some or all of the functions of the elements of the processing system 100. In addition, some or all of the functions of the elements of the processing system 100 (e.g., engines 110, 112, 114, 116) can be implemented as nodes of the cloud computing system. For example, the ML model training engine 110 is implemented in the cloud computing system so that the training of the ML model is performed in the cloud computing system. The model can then be transmitted to the processing system 100 or otherwise made available to the processing system, such as via a network.

[0031] As described herein, the machine learning model can be trained to perform real-time (or near real-time) distortion prediction for additive manufacturing, which is useful for improving the additive manufacturing process. More specifically, the present technology can combine and utilize rule-based decision making and artificial intelligence (AI) reasoning to complete the various operations described herein, namely, distortion prediction for additive manufacturing. The phrase "machine learning" broadly describes the function of an electronic system that learns from data. A machine learning system, engine, or module may include a trainable machine learning algorithm that can be trained, such as in an external cloud environment, to learn the functional relationship between currently unknown inputs and outputs, and the resulting model can be used to perform segmentation of CT voxel data. In one or more embodiments, an artificial neural network (ANN) with the ability to be trained to perform currently unknown functions can be used to implement machine learning functions. In machine learning and cognitive science, ANN is a family of statistical learning models inspired by the biological neural networks of animals (and specifically the brain). ANN can be used to evaluate or approximate systems and functions that rely on a large number of inputs. Convolutional neural networks (CNNs) are a class of deep feedforward ANNs that are particularly useful in analyzing visual images. A recurrent neural network (RNN) is a class of artificial neural networks that operate on sequential or time series data. RNNs are trained using training data. RNNs learn from previous inputs using a feedback loop that acts as a "memory" for storing the state or information of previous inputs for generating the next output. Examples of architecture types for RNNs include long-term memory (LSTM), bidirectional RNN, or gated recurrent unit.

[0032] ANNs may be embodied as so-called "neuromorphic" systems of interconnected processor elements that act as simulated "neurons" and exchange "messages" between each other in the form of electronic signals. Similar to the so-called "plasticity" of synaptic neurotransmitter connections that carry messages between biological neurons, the connections in ANNs that carry electronic messages between simulated neurons are provided with numerical weights corresponding to the strength or weakness of a given connection. The weights may be adjusted and tuned based on experience so that the ANN adapts to the input and is able to learn. For example, an ANN for handwriting recognition is defined by a set of input neurons that may be activated by pixels of an input image. After being weighted and transformed by a function determined by the designer of the network, the activation of these input neurons is then passed to other downstream neurons, which are generally referred to as "hidden" neurons. The process is repeated until the output neuron is activated. The activated output neuron determines which character is read. It should be understood that these same techniques may be applied to situations where real-time (or near real-time) distortion prediction is performed for additive manufacturing.

[0033] The ML model training engine 110 trains a machine learning model, such as a recurrent neural network, using training data 120 stored in a data store 108 or another suitable device. The ML model inference engine 112 uses the trained ML model to make predictions about distortion. Inference may be supplemented with one or more of image analysis performed by an image analysis engine 114 and / or construction simulation performed by a construction simulation engine 116.

[0034] Using engines 110, 112, 114, 116, the processing system 100 uses image analysis to predict the distortion of additive manufacturing. One embodiment is described as follows. For example, the processing system 100 performs a pre-built simulation to generate simulated distortion. The processing system 100 also performs image analysis on the camera image to detect the real distortion (e.g., exposed area and interference) on the layer. The results of the simulated distortion and image analysis are applied to the trained machine learning model (e.g., recursive neural network) to evaluate the distortion in the current layer and predict the expected distortion in the next layer. This provides real-time (or near real-time) evaluation of dimensional inaccuracy and / or predicts recoater interference, which provides corrective actions such as changing laser parameters or paths and / or skipping the next layer in the selected area to minimize distortion. Additional examples of corrective actions include using a larger layer thickness to avoid interference, increasing the time interval between subsequent layers to cool the part, not further printing the part but continuing to print other parts on the board, continuing as is, stopping and abandoning the construction, etc., including their combinations and / or multiples. In some cases, the operator of the additive manufacturing system can be notified, and they can implement appropriate actions. In addition, the results can be used to improve the design and / or construction settings for future manufacturing. In some examples, the results can be used to retrain the machine learning model. It should be understood that one or more other embodiments are also possible. Figure 2 to- Figure 8 The features and functionality of engines 110, 112, 114, 116 are described in more detail.

[0035] Figure 2 A flow chart of a method 200 for predicting distortion for additive manufacturing using image analysis according to one or more embodiments described herein is shown. The method 200 may be implemented using any suitable system and / or device. For example, the method 200 may be implemented using Figure 1 A processing system 100, Fig. 9 The invention may be implemented by processing system 900, etc. (including combinations and / or multiples thereof).

[0036] At box 202, an image 203 is received or captured (such as by sensor 106). Image 203 is an image of a layer of an object manufactured by an additive manufacturing system. At box 204, an image of a fixed z height relative to a build volume of the additive manufacturing system is taken, producing a slice stack 205. The slice stack represents build instructions for manufacturing a layer (corresponding to image 203) by the additive manufacturing system.

[0037] At block 206, processing system 100 performs image analysis on image 203 using image analysis engine 114. Specifically, image analysis engine 114 performs image analysis on image 203 to identify exposed surfaces 213 in image 203 of the layer. Exposed surfaces represent areas of the layer where material has been removed or otherwise disturbed, which may be caused, for example, by a recoater. The exposed surfaces in image 203 from the current layer (and / or from a previous layer) are then compared to simulated distortions generated by build simulation engine 116. That is, build simulation engine 116 generates simulated distortions of the layer (or layers), as further described herein, and compares the simulated distortions to the results of the image analysis.

[0038] At block 208, the trained machine learning model is applied to the comparison results at block 206 to predict the distortion of the next layer. At block 210, inference is performed (e.g., recoater inference, dimensional inspection, etc., including combinations and / or multiples thereof).

[0039] Additional processes may also be included, and it should be understood that Figure 2 The processes depicted in the drawings represent examples, and other processes may be added or existing processes may be removed, modified, or rearranged without departing from the scope of the present disclosure.

[0040] Now go to Figure 3A to- Figure 3F, image analysis is now described. For example, images 301-306 are shown for image analysis using the image analysis engine 114 according to one or more embodiments described herein. Specifically, Figure 3A-3D Images 301-304 of layers of an object manufactured by an additive manufacturing system according to one or more embodiments described herein are shown. Images 301-304 show post-recoating images taken during the construction of four different layers (layers 395, 445, 566, and 791, respectively). These layers are used as examples only, and the techniques described herein can be applied to any layer and / or any number of layers. At layer 395 (image 301), the exposed area 311 of the part is not covered by powder and exposed due to out-of-plane deformation (e.g., curling or ridges) after recoating. This is observed as a bright or shiny object in image 301. When the deformed part collides with the recoater blade (called interference), the exposed area 311 grows through layer 566 with curling growth (image 302), and causes inappropriate spreading characterized by stripes 312 (images 303, 304). For additional layers, interference continues and eventually disappears as curling decreases. The streak is still visible because the recoater blade is damaged by that spot and cannot spread the powder properly.In accordance with one or more embodiments described herein, any other type of image (eg, thermal image) may be used.

[0041] The processing system 100 uses the image analysis engine 114 to perform image analysis on the images 301-304. Such image analysis may include, for example, image processing, computer vision algorithms, and the like, including combinations and / or multiples thereof. Image analysis may be used to detect exposed areas, streaks, and / or recoater interference, etc., from the post-recoating images in real time (or near real time). Images 305 and 306 show detection areas 321, 322 corresponding to the exposed areas 311 and streaks 312 of the images 301-304, which the processing system 100 identifies using, for example, a computer vision segmentation model. Specifically, according to one or more embodiments described herein, Figure 3E An image 305 of an exposed area 321 is shown, and Figure 3F Image 306 of streaks 322 caused by recoater interference is shown.

[0042] Now go to Figure 4A-4D , the construction simulation is now described. For example, images 401-404 show simulated deformations of layers 395, 445, 566, and 791, respectively (see, e.g. Figure 3A-3D ). Images 401-404 are layer-by-layer deformations predicted by simulation. The processing system 100 predicted the deformations shown in images 401-404 using the build simulation engine 116. The build simulation engine 116 performs a physics-based simulation of the build process. Figure 4A-4DThe color contours of the are shown as out-of-plane deformations (curls). For example, when the predicted curl exceeds a certain threshold value that depends on the thickness of the powder layer, a recoater disturbance is predicted. The simulation may take into account one or more of the planned material type, processing parameters, etc., but in at least one example, not necessarily all phenomena, unknown physics, and / or actual conditions during the build are considered, which may be different from those planned / expected. Therefore, actual deformations and disturbances may not always be consistent with the predictions.

[0043] Build data may also be used during distortion prediction. For example, the processing system 100 may store build data (BD) 122, such as in the data storage 108. For example, a large number of factors, such as raw material (powder), design, build settings, chamber conditions, processing parameters, etc. (including combinations thereof) may affect deformation during build (i.e., manufacturing). The build simulation may not take all of these factors into account. In addition, the actual values ​​of these factors may be different from the build plan. These factors may be static factors and / or dynamic factors. Static factors are substantially constant for the entire build, such as powder size distribution (PSD), powder morphology, etc., including combinations thereof. During the build, dynamic factors of the layer may change, such as moisture content, laser power, scanning speed, airflow, etc., including combinations thereof.

[0044] Now go to Figure 5A and Figure 5B , schematic diagrams 501, 502 of spreading simulations according to one or more embodiments described herein are shown. Factors such as PSD, powder morphology, chamber conditions, recoater type, recoater speed, damage to the recoater, layer thickness, deformed geometry, etc. (including combinations thereof) can affect the spreading behavior of the powder during the build. The build simulation engine 116 can perform a spreading simulation that models the spreading of the powder on the deformed geometry taking into account one or more of these factors. For example, the spreading simulation generates predicted exposed areas and / or recoater interference. Figure 5A Schematic diagram 501 shows an exposed area 511 of a deformed part 512 for large-sized particles 513. Figure 5B Schematic 502 shows the exposed area 511 of a deformed feature 512 for a large size particle 523. By comparing the two schematics, it can be observed that the same amount of curling can result in significantly different amounts of exposed area depending on the particle size distribution. Similarly, for the same powder size, different geometries can result in disproportionately different amounts of exposed area (not shown in the figure). The spreading simulation takes these factors into account.

[0045] Now describe Figure 6 and Figure 7, which provides a method for distortion prediction for additive manufacturing using image analysis according to one or more embodiments described herein. Preferably, Figure 6 A flow chart of a method 600 for predicting distortion for additive manufacturing using image analysis according to one or more embodiments described herein is shown. The method 600 may be implemented using any suitable system and / or device. For example, the method 600 may be implemented using Figure 1 A processing system 100, Fig. 9 The invention may be implemented by processing system 900, etc. (including combinations and / or multiples thereof).

[0046] Method 600 uses image analysis (box 602), build simulation (box 604), and build data (box 606) as inputs to a trained machine learning model (box 608). The variable "n" is the current layer number, (n+1) is the next layer to be printed, and k is the number of previous layers considered in the model (e.g., one or more predictions may depend on the previous k layers). According to one or more embodiments described herein, a build simulation (box 604) is performed before the build begins so that the results are available for each of the layers. For example, the trained machine learning model (box 608) makes one or more predictions of deformation for the current layer "n" and / or the next layer "n+1" (box 610). Based on the predicted deformation, the exposed area and recoater interference of the current layer "n" and / or the next layer "n+1" may be predicted using appropriate criteria or calculations (box 612). Influencing factors may include one or more of PSD, layer thickness, melting effects, etc., including combinations thereof. When predicting a recoater disturbance, the severity of the recoater disturbance and / or the location of the recoater disturbance may also be predicted.

[0047] Additional processes may also be included, and it should be understood that Figure 6 The processes shown in represent examples, and other processes may be added or existing processes may be removed, modified, or rearranged without departing from the scope of the present disclosure.

[0048] Figure 7 A flow chart of a method for predicting distortion for additive manufacturing using image analysis according to one or more embodiments described herein is shown. The method 700 may be implemented using any suitable system and / or device. For example, the method 700 may be implemented using Figure 1 A processing system 100, Fig. 9 The invention may be implemented by processing system 900, etc. (including combinations and / or multiples thereof).

[0049] Method 700 uses image analysis (box 702), spread simulation (box 704), and build data (box 708) as inputs to a trained machine learning model (box 710). The spread simulation (box 704) uses the deformation predicted by the build simulation (box 708) as input and output for predicting exposed areas and / or recoater interference. The trained machine learning model (box 710) predicts exposed areas and recoater interference for the current layer "n" and / or the next layer "n+1" (box 712).

[0050] Additional processes may also be included, and it should be understood that Figure 7 The processes shown in the drawings represent examples, and other processes may be added or existing processes may be removed, modified, or rearranged without departing from the scope of the present disclosure.

[0051] Figure 8 A flow chart of a method 800 for training a machine learning model according to one or more embodiments described herein is shown. The method 800 may be implemented using any suitable system and / or device. For example, the method 800 may be implemented using Figure 1 A processing system 100, Fig. 9 The invention may be implemented by processing system 900, etc. (including combinations and / or multiples thereof).

[0052] At box 802, method 800 begins. At box 804, an untrained machine learning model is initialized. At box 806, training data (e.g., training data 120) is received (such as from another system or device), collected (such as using sensor 106), and / or read (such as from data storage 108). Examples of training data include past builds from an additive manufacturing system where (i) no areas were exposed, (ii) areas were exposed but recoater interference did not occur, and (iii) exposed areas that caused recoater interference. Because the builds are from the past, for the next layer from the image (layer "n+1"), the "ground truth" (e.g., actual exposed areas and interferences in box 810) is known. The predictions from box 808 (see, for example, respectively, Figure 6 and Figure 7 Increasing the number of past constructions that represent various situations in the training data increases the fidelity of the trained machine learning model.

[0053] At block 808, the processing system 100 begins training by running the machine learning model on the training data using the ML model training engine 110. The ML model training engine 110 generates a prediction of the exposed area and / or the recoater interference. At block 810, the prediction generated at block 808 is compared to the actual exposed area and / or the actual recoater blade interference of the additive manufacturing system. That is, at block 810, the ML model training engine 110 compares at least one of the predicted exposed area and the predicted recoater interference to the actual exposed area or the actual predicted recoater interference.

[0054] At decision block 812, the ML model training engine 110 determines whether the accuracy of the prediction from block 808 is acceptable based on the comparison at block 810. For example, if the prediction is within a threshold deviation of the actual exposure area and the recoater blade interference (e.g., within 10%, within 5%, within 2.5%, within 2%, within 1%, within 0.1%, etc.), the prediction is considered acceptable. In such cases, the method 800 proceeds to block 814, and the trained ML model is generated, saved, and / or output so that it can be used for inference as described herein. However, if it is determined at decision block 812 that the accuracy of the prediction from block 808 is unacceptable, the method 800 proceeds to block 816, where the ML model parameters are adjusted for the next iteration. Examples of such ML model parameters include weights of one or more layers (e.g., hidden layers, output layers, etc., including combinations thereof) of the machine learning model. Training (e.g., blocks 808, 810, 812, 814) may be iterated one or more times until the ML model is trained.

[0055] Additional processes may also be included, and it should be understood that Figure 8 The processes shown in represent examples, and other processes may be added or existing processes may be removed, modified, or rearranged without departing from the scope of the present disclosure.

[0056] It should be understood that one or more of the embodiments described herein can be implemented in conjunction with any other suitable type of computing environment now known or later developed. Fig. 9A block diagram of a processing system 900 is depicted, which can be used to implement the techniques described herein. According to one or more embodiments described herein, the processing system 900 is an example of a cloud computing node of a cloud computing environment. In the example, the processing system 900 has one or more central processing units ("processors" or "processing resources" or "processing devices") 921a, 921b, 921c, etc. (collectively or collectively referred to as processors 921 and / or processing devices). In various aspects of the present disclosure, each processor 921 may include a reduced instruction set computer (RISC) microprocessor. The processor 921 is coupled to a system memory (e.g., a random access memory (RAM) 924) and various other components via a system bus 933. A read-only memory (ROM) 922 is coupled to the system bus 933 and may include a basic input / output system (BIOS) that controls certain basic functions of the processing system 900.

[0057] Also shown are an input / output (I / O) adapter 927 and a network adapter 926 coupled to the system bus 933. The I / O adapter 927 may be a small computer system interface (SCSI) adapter that communicates with the hard disk 923 and / or the storage device 925 or any other similar component. The I / O adapter 927, the hard disk 923, and the storage device 925 are collectively referred to herein as a mass storage device 934. An operating system 940 for execution on the processing system 900 may be stored in the mass storage device 934. The network adapter 926 interconnects the system bus 933 with an external network 936, enabling the processing system 900 to communicate with other such systems.

[0058] A display (e.g., a display monitor) 935 is connected to the system bus 933 via a display adapter 932, which may include a graphics adapter and a video controller for improving the performance of graphics-intensive applications. In one aspect of the present disclosure, adapters 926, 927, and / or 932 may be connected to one or more I / O buses that are connected to the system bus 933 via an intermediate bus bridge (not shown). Suitable I / O buses for connecting peripheral devices such as hard disk controllers, network adapters, and graphics adapters typically include a common protocol, such as a peripheral component interconnect (PCI). Additional input / output devices are shown as being connected to the system bus 933 via a user interface adapter 928 and a display adapter 932. A keyboard 929, a mouse 930, and a speaker 931 may be interconnected to the system bus 933 via a user interface adapter 928, which may include, for example, a super I / O chip that integrates multiple device adapters into a single integrated circuit.

[0059] In some aspects of the present disclosure, the processing system 900 includes a graphics processing unit 937. The graphics processing unit 937 is a specialized electronic circuit designed to manipulate and alter memory to speed up the creation of images in a frame buffer intended for output to a display. Generally speaking, the graphics processing unit 937 is very efficient in manipulating computer graphics and image processing, and has a highly parallel structure, making it more efficient than a general-purpose CPU for algorithms that complete large block processing in parallel.

[0060] Thus, as configured herein, the processing system 900 includes processing capabilities in the form of a processor 921, storage capabilities including system memory (e.g., RAM 924) and mass storage 934, input devices such as a keyboard 929 and a mouse 930, and output capabilities including speakers 931 and a display 935. In some aspects of the present disclosure, a portion of the system memory (e.g., RAM 924) and mass storage 934 together store an operating system 940 to coordinate the functionality of the various components shown in the processing system 900.

[0061] Some embodiments of the aforementioned disclosure are shown below:

[0062] Embodiment 1: A method includes: performing image analysis on an image of a layer of an object manufactured by an additive manufacturing system to identify exposed surfaces in the image of the layer; performing a build simulation to generate a simulated distortion of the layer; evaluating build data to determine a value of an impact factor for the layer; using a machine learning model to predict at least one of a predicted distortion or a predicted recoater interference for a next layer based at least in part on the image analysis, the build simulation, and the build data; and implementing an action to change the manufacture of the next layer based at least in part on at least one of the predicted distortion or the predicted recoater interference.

[0063] Embodiment 2: According to the method of any one of the preceding embodiments, the method further comprises performing a spreading simulation.

[0064] Embodiment 3: A method according to any one of the preceding embodiments, wherein predicting the prediction distortion of the next layer is also based at least in part on the results of the spreading simulation.

[0065] Embodiment 4: A method according to any of the preceding embodiments, wherein the spreading simulation models the spreading of the powder on the layer based at least in part on at least one factor, wherein the at least one factor is selected from powder size distribution, powder morphology, chamber conditions, recoater type, recoater speed, damage on the recoater, layer thickness and deformed geometry.

[0066] Embodiment 5: The method according to any one of the preceding embodiments, wherein predicting the predicted recoater interference includes predicting a severity of the predicted recoater interference and a location of the predicted recoater interference.

[0067] Embodiment 6: A method according to any of the preceding embodiments, wherein the machine learning model is a recursive neural network.

[0068] Embodiment 7: According to the method described in any of the preceding embodiments, the method also includes training the machine learning model.

[0069] Embodiment 8: The method according to any of the preceding embodiments, wherein training the machine learning model comprises inputting training data into the machine learning model to generate at least one of a predicted exposure area or a predicted recoater interference.

[0070] Embodiment 9: A method according to any of the preceding embodiments, wherein training the machine learning model further comprises comparing at least one of the predicted exposure area and the predicted recoater interference with the actual exposure area or the actual predicted recoater interference.

[0071] Embodiment 10: A method according to any of the preceding embodiments, wherein training the machine learning model further comprises generating a trained machine learning model in response to determining that the accuracy of the comparison is acceptable.

[0072] Embodiment 11: A method according to any of the preceding embodiments, wherein training the machine learning model further includes performing at least one additional training iteration in response to determining that the accuracy of the comparison is unacceptable, wherein at least one machine learning model parameter is adjusted during each iteration.

[0073] Embodiment 12: A processing system comprising a memory and a processor, the processing system being used to execute computer-readable instructions, the computer-readable instructions controlling the processor to perform operations, the operations comprising: performing image analysis on an image of a layer of an object manufactured by an additive manufacturing system to identify exposed surfaces in the image of the layer; performing a build simulation to generate a simulated distortion of the layer; comparing the exposed surfaces in the image of the layer with the simulated distortion of the layer; predicting the distortion of a next layer using a machine learning model; and implementing actions based at least in part on the predicted distortion to reduce distortion during manufacture of the next layer.

[0074] Embodiment 13: A system according to any of the preceding embodiments, wherein the machine learning model is a recursive neural network.

[0075] Embodiment 14: According to any of the preceding embodiments, the system also includes the additive manufacturing system.

[0076] Embodiment 15: According to the system of any of the preceding embodiments, the instructions also include training the machine learning model, wherein training the machine learning model includes inputting training data into the machine learning model to generate at least one of a predicted exposure area or a predicted recoater interference.

[0077] Embodiment 16: A system according to any of the preceding embodiments, wherein training the machine learning model further comprises comparing at least one of the predicted exposure area and the predicted recoater interference with the actual exposure area or the actual predicted recoater interference.

[0078] Embodiment 17: A system according to any of the preceding embodiments, wherein training the machine learning model further includes generating a trained machine learning model in response to determining that the accuracy of the comparison is acceptable.

[0079] Embodiment 18: A system according to any of the preceding embodiments, wherein training the machine learning model further includes performing at least one additional training iteration in response to determining that the accuracy of the comparison is unacceptable, wherein at least one machine learning model parameter is adjusted during each iteration.

[0080] Embodiment 19: A system according to any of the preceding embodiments, the system also includes a camera, wherein the operation also includes capturing images of layers of the object manufactured by the additive manufacturing system during the manufacturing.

[0081] Embodiment 20: The system of any of the preceding embodiments, wherein performing the action comprises at least one of changing a laser power of the additive manufacturing system or skipping manufacturing of at least a portion of the next layer.

[0082] In the context of describing the present disclosure (particularly in the context of the appended claims), the use of the terms "a," "an," and "the," and similar references should be construed to cover both the singular and the plural, unless otherwise indicated herein or clearly contradicted by context. In addition, it should be noted that the terms "first," "second," and the like, herein do not denote any order, quantity, or importance, but are used to distinguish one element from another. The modifier "about" used in conjunction with a quantity is inclusive of the stated value and has a meaning dictated by the context (e.g., it includes the degree of error associated with the measurement of the particular quantity).

[0083] Although the present disclosure has been described with reference to one or more exemplary embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted for its elements without departing from the scope of the present disclosure. In addition, many modifications may be made to adapt specific situations or materials to the teachings of the present disclosure without departing from the basic scope of the present disclosure. Therefore, it is contemplated that the present disclosure is not limited to the specific embodiments disclosed as the best mode for implementing the present disclosure, but that the present disclosure will include all embodiments falling within the scope of the claims. In addition, in the drawings and specific embodiments, exemplary embodiments of the present disclosure have been disclosed, and although specific terms have been adopted, unless otherwise indicated, they are used only in a general and descriptive sense, and not for the purpose of limitation, and the scope of the present disclosure is therefore not limited thereto.

Claims

1. A method (600), comprising: performing image analysis (602) on an image (203) of a layer of an object manufactured by an additive manufacturing system to identify exposed surfaces (213) in the image of the layer; performing a build simulation (604) to generate a simulated distortion of the layer; evaluating the build data (606) to determine a value for the impact factor of the layer; predicting at least one of a predicted distortion or a predicted recoater interference for a next layer (610) using a machine learning model (608) based at least in part on the image analysis (602), the build simulation (604), and the build data (606); and Based at least in part on at least one of the predicted distortion or the predicted recoater disturbance (610), an action is performed to change the fabrication of the next layer.

2. The method (600) of claim 1, further comprising performing a spreading simulation (704).

3. The method (600) of claim 2, wherein predicting the prediction distortion (610) of the next layer is also based at least in part on results of the spreading simulation (704).

4. The method (600) of claim 2, wherein the spreading simulation (704) models the spreading of the powder on the layer based at least in part on at least one factor, wherein the at least one factor is selected from powder size distribution, powder morphology, chamber conditions, recoater type, recoater speed, damage on the recoater, layer thickness, and deformed geometry.

5. The method (600) of claim 1, wherein predicting the predicted recoater interference (610) comprises predicting a severity of the predicted recoater interference and a location of the predicted recoater interference.

6. The method (600) of claim 1, wherein the machine learning model (608) is a recurrent neural network.

7. The method (600) of claim 1, further comprising training the machine learning model (608).

8. The method (600) of claim 7, wherein training the machine learning model (608) comprises inputting training data into the machine learning model to generate at least one of a predicted exposure area or a predicted recoater interference (610).

9. The method (600) of claim 8, wherein training the machine learning model (608) further comprises comparing the predicted exposure area and at least one of the predicted recoater interference (610) with an actual exposure area or an actual predicted recoater interference.

10. The method (600) of claim 9, wherein training the machine learning model (608) further comprises generating a trained machine learning model (608) in response to determining that the accuracy of the comparison is acceptable.

11. The method (600) of claim 9, wherein training the machine learning model (608) further comprises, in response to determining that the accuracy of the comparison is unacceptable, performing at least one additional training iteration, wherein at least one machine learning model parameter is adjusted during each iteration.

12. A system, comprising: A processing system (100) comprising a memory (104) and a processing device (102), the processing system being configured to execute computer-readable instructions, the computer-readable instructions controlling the processing device (102) to perform operations, the operations comprising: performing image analysis (602) on an image (203) of a layer of an object manufactured by an additive manufacturing system to identify exposed surfaces (213) in the image of the layer; performing a build simulation (604) to generate a simulated distortion of the layer; comparing the exposed surface (213) in the image of the layer to a simulated distortion of the layer; predicting the distortion (610) of the next layer using the machine learning model (608); and Based at least in part on the predicted distortion, actions are performed to reduce distortion during fabrication of the next layer.

13. The system of claim 12, wherein the machine learning model (608) is a recurrent neural network.

14. The system of claim 12, further comprising the additive manufacturing system.

15. The system of claim 14, wherein the instructions further comprise training the machine learning model (608), wherein training the machine learning model (608) comprises inputting training data into the machine learning model (608) to generate at least one of a predicted exposure area or a predicted recoater interference.

16. The system of claim 15, wherein training the machine learning model (608) further comprises comparing at least one of the predicted exposure area and the predicted recoater interference with an actual exposure area or an actual predicted recoater interference.

17. The system of claim 16, wherein training the machine learning model (608) further comprises generating a trained machine learning model (608) in response to determining that the accuracy of the comparison is acceptable.

18. The system of claim 16, wherein training the machine learning model (608) further comprises, in response to determining that the accuracy of the comparison is unacceptable, performing at least one additional training iteration, wherein at least one machine learning model parameter is adjusted during each iteration.

19. The system of claim 12, further comprising a camera (106), wherein the operations further comprise capturing the image (203) of the layer of the object manufactured by the additive manufacturing system during the manufacturing.

20. The system of claim 12, wherein performing the action comprises at least one of changing a laser power of the additive manufacturing system or skipping manufacturing of at least a portion of the next layer.

Images