Machine learning with fast feature generation

Abstract

A method comprising: identifying machine process parameters of an additive manufacturing process used to produce a part; providing a real-world sensor to sense characteristics associated with a real-world version of the additive manufacturing process; receiving sensor readings from the real-world sensor while a machine executes the real-world version of the additive manufacturing process; generating, using a computer-based processor, physics-based features associated with the additive manufacturing process; and training a machine learning software model based at least in part on the machine process parameters, the sensor readings, and the physics-based features to predict performance of the real-world sensor.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims the benefit of priority to U.S. Provisional Patent Application No. 62 / 837,452, filed on April 23, 2019, entitled “MACHINE LEARNING SERVICE BASED ON V+R DATA” and claims the benefit of priority to U.S. Provisional Patent Application No. 62 / 838,114, filed on April 23, 2019, entitled “MACHINE LEARNING WITH FAST FEATURE GENERATION FOR SELECTIVE LASER MELTING PRINT PARAMETERS OPTIMIZATION”. The entire disclosures of these prior applications are incorporated herein by reference. Technical Field

[0003] The present application relates to machine learning, and more particularly, to machine learning based on virtual data and real data. Background Art

[0004] In hardware, sensor values ​​are used for control (e.g., aircraft flight control or manufacturing control). Today, cyber-physical control systems are often designed using virtual models. Therefore, it is important to accurately predict sensor values. In addition, manufacturing is challenged to produce product parts of acceptable quality at an early stage (e.g., the first time).

[0005] In particular, the complexities associated with additive manufacturing processes include a large number of printing parameters and rapidly evolving physical properties occurring at different scales, such as material phase transformations and porosity at the melt pool level, as well as distortion and residual stress development at the part level. It is often costly to use physical simulations to predict all quality metrics. Summary of the invention

[0006] In one aspect, a method includes: identifying machine process parameters of an additive manufacturing process used to produce a part; providing a real-world sensor to sense characteristics associated with a real-world version of the additive manufacturing process; receiving sensor readings from the real-world sensor while a machine executes the real-world version of the additive manufacturing process; generating, using a computer-based processor, physics-based features associated with the additive manufacturing process; and training a machine learning software model based at least in part on the machine process parameters, the sensor readings, and the physics-based features to predict performance of the real-world sensor.

[0007] In certain embodiments, the physics-based features include an energy flux feature and an effective conduction feature. The energy flux feature generally corresponds to the amount of energy delivered to the region of interest during a specific period of time during the additive manufacturing process. The effective conduction feature generally corresponds to the ability to dissipate energy introduced into the region of interest through thermal conduction.

[0008] In another aspect, a system includes a computer, the computer including: a computer-based processor; and a computer-based memory storing computer-executable instructions that, when executed by the computer-based processor, cause the computer-based processor to generate physics-based features associated with an additive manufacturing process. The system includes: one or more machines configured to execute a real-world version of the additive manufacturing process based on machine process parameters associated with the additive manufacturing process; real-world sensors to sense characteristics associated with the real-world version of the additive manufacturing process; and a machine learning software model trained to predict performance of the real-world sensors based on the physics-based features, the machine process parameters, and sensor readings from the real-world sensors in the real-world version of the additive manufacturing process.

[0009] In certain embodiments, one or more of the following advantages are present.

[0010] For example, highly accurate sensor predictions are possible. Furthermore, predictions can be achieved without the need to model the specific operation of the sensor from first principles and without extensive calibration efforts. Highly accurate sensor predictions can be achieved efficiently. Once trained, the prediction model can be used in a variety of ways, including: to optimize machine process parameters, to detect sensor anomalies, and / or to support subsequent system simulations.

[0011] In addition, in particular in certain embodiments, the disclosed techniques provide for rapid training and rapid prediction of sensor performance, which is particularly useful in situations where available time is limited but accuracy and precision are required.

[0012] In addition, in some embodiments, the systems and techniques disclosed in the present invention provide easy-to-use tools that can help quickly obtain the printing quality of any new design parts. In addition, in some embodiments, without involving complex simulation analysis to model complex processes, the systems and techniques disclosed in the present invention can be used to directly generate optimized machine parameters in a machine-readable format, so that, for example, optimized printing parameters can be obtained.

[0013] Other features and advantages will be apparent from the description and drawings, and from the claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a schematic diagram of a method that includes training a machine learning software model 106 using real data and virtual data about a process to be able to predict the behavior of a sensor in a real version of the process.

[0015] Figure 2 yes Figure 1 Schematic diagram of a more detailed representation of the method shown.

[0016] Figure 3 is a schematic cross-sectional view of an exemplary three-dimensional (3D) printer.

[0017] Figure 4 yes Figure 2 Schematic diagram of an implementation of the method represented in a 3D printing environment specific manner.

[0018] Figure 5 Several examples of computer-aided design (CAD) models with different geometries are shown.

[0019] Figures 6 to 9 Shows the Screen shot of the Powder Bed Manufacturing application which is part of the software platform.

[0020] Fig.10 A schematic diagram showing a top view of a powder bed in which a part is being printed in a 3D printer and the path that a laser tool will travel along the upper surface of the powder bed in rectangular print layers.

[0021] Fig.11 and Fig.12 yes Fig.10 Schematic diagram of the powder bed, showing the powder bed at the beginning of the scanning process (Figure 11) and at the end of the scanning process ( Fig.12 Laser position and area of ​​interest circles in the middle.

[0022] Fig.13 Shown is a plot of the predicted scatter for a single layer of test data without including physics-based features during training.

[0023] Fig.14 Shown is a plot of predicted scatter for a single layer of test data including physics-based features during training.

[0024] Fig.15 Experimental validation of photodiode sensor readings for a single layer before and after the printing optimization process is shown.

[0025] Fig.16 is a schematic diagram showing an exemplary flow of a machine learning and print optimization framework.

[0026] Fig.17 Schematic diagram of the process used to perform sensor prediction and optimization of machine parameters layer by layer and in real time in an additive manufacturing environment.

[0027] Fig.18 is a schematic diagram illustrating an exemplary flow of sensor abnormality detection.

[0028] Fig.19 A schematic diagram of a computer network or similar digital processing environment.

[0029] Fig. 20 yes Fig.19 A schematic diagram of an exemplary computer in a computer network or similar digital processing environment.

[0030] Like reference characters denote like elements. DETAILED DESCRIPTION

[0031] The present application relates to systems and processes related to predicting the behavior of sensors in a process performed by a real-world machine or system (hereinafter referred to as a "machine"). The process can be a manufacturing process or any other type of process that causes at least one sensor to sense a feature associated with the process. The machine can be any machine or system configured to perform the process and including a sensor. For example, in the additive manufacturing industry, three-dimensional (3D) printers typically include photodiode sensors to sense light at or near the 3D printer molten pool. The light level measured by the photodiode sensor is related to the heat transferred to the molten pool. The 3D printing industry typically regards data collected by photodiode sensors in 3D printers as part quality indicators of printed parts. Other examples of processes to which the systems and methods disclosed herein can be applied include autonomous driving of vehicles, spacecraft processes, and many other processes.

[0032] Figure 1 is a schematic diagram of a method that includes training a machine learning software model 106 using real data 102 and virtual data 104 about a process to be able to predict the behavior of sensors in a real version of the process. Figure 1, the machine learning software model 106 can be applied to any of a variety of possible uses, including, for example, controlling / optimizing the machine (at 108) to perform the same type of process or a different type of process, diagnosing the machine and / or sensors in the machine (at 110), and / or simulating various aspects of the behavior of the machine / sensor or process under various operating parameters (at 112). Of course, other uses are possible. Moreover, in some embodiments, the machine learning software model 106 can be used in one or more of these processes while continuing to be trained. In those cases, continued training will continue to evolve the machine learning software model 106 to even better predict sensor behavior.

[0033] In typical embodiments, the systems and methods disclosed herein (e.g., Figure 1 ) results in highly accurate sensor predictions. Moreover, in certain embodiments, these predictions can be made very quickly, which is particularly important in situations where limited time is available for extensive training of the machine learning software model 106. In these situations, highly accurate predictions can be achieved despite the limited time available during real-time processes. This makes the systems and methods disclosed herein very valuable in a variety of different applications.

[0034] The real data 102 utilized to train the machine learning software model 106 can be any type of data from a real-world version of a process. For the real data 102, there are a variety of possible sources.

[0035] One possible source of real data 102 is a real-world version of a sensor whose behavior the machine learning software model 106 is learning to predict. Typically, such a real-world sensor will be deployed in an operating machine to sense one or more conditions produced by the machine while performing the process in question. For example, in an additive manufacturing setting such as 3D printing, the machine learning software model 106 may be trained to predict the behavior of a photodiode sensor in a 3D printer, which is configured to sense light in the 3D printer (as an indicator of thermal energy being delivered to an object to be printed). In those embodiments, the real data may include actual data collected by the photodiode sensor during the 3D printing process.

[0036] Another possible source of real data 102 is process parameters or settings used on a real machine to implement the process. In the 3D printing example above, process parameters may include information or settings on the 3D printer related to one or more of the following: laser exposure time, point-to-point distance, laser power, laser speed, and laser strategies including contour and hatching / fill patterns, radial step, offset, direction and sequence, etc. Certain process parameters or machine settings may be relevant to training the machine learning software model 106 to predict the behavior of the photodiode sensor in the 3D printer example because they affect the amount of heat transferred into the printed part. Other process parameters in a 3D printing environment or other types of processes may also be relevant.

[0037] The real data 102 may also come from other sources.

[0038] The virtual data 104 used to train the machine learning software model 106 typically comes from simulation or virtual modeling of one or more aspects of the process or machine in question.

[0039] The simulations that generate the virtual data 104 are typically performed by a computer-based simulator and can be relatively simple or refined. Simpler simulations tend to be faster, which may be desirable in some cases, but contain fewer details. Complex and refined simulations provide more details about the interaction of physical phenomena, but tend to take more time. In general, the simulation method should be adapted to the requirements of a particular situation.

[0040] For example, in a relatively simple simulation, the simulator may calculate virtual data 104 in the form of one or more physics-based features (e.g., features related to the underlying physics of the process in question) based on readily available process information, such as one or more of the process parameters mentioned above. Physics-based features are generally not directly measurable and may not be easily measured in an accurate manner by physical sensors. Some examples of physics-based features suitable for use in a 3D printing environment include some measure of the energy introduced into a part being printed within a specific time window, and some measure of the part's ability to dissipate the introduced energy by thermal conduction.

[0041] More complex simulations can be performed with sophisticated computer-based simulators, such as the Powder Bed Manufacturing and Additive Manufacturing Solutions software applications, which are part of Dassault Systèmes SE. part of the software platform provided. The powder bed manufacturing software application is capable of generating virtual data 104 in the form of high-resolution machine tool path information based on a computer-aided design (CAD) model and, in some cases, other input data. The high-resolution machine tool path information generated in this regard may include information similar to the process parameters described above, for example, information related to one or more of the following: laser exposure time, point-to-point distance, laser power, laser speed, laser strategy including contour and hatching / fill patterns, radial step size, offset, direction and sequence, etc. The additive manufacturing solution software application is capable of generating virtual data 104 in the form of high-fidelity physics information based on a computer-aided design (CAD) model machine tool path information and, in some cases, other input data. The high-fidelity physics information generated in this regard may include virtual data of physical quantities such as temperature field, melt pool size, material phase change and content, mechanical deformation, residual stress and strain, etc.

[0042] In some implementations, virtual data 104 may include a combination of physics-based features and information from more complex computer-based simulators.

[0043] Real data 102 and virtual data 104 are typically collected, for example, in a computer-based memory, and used with a trained machine learning software model 106 to be able to predict the real behavior of sensors (e.g., a photodiode sensor in a 3D printer) for various computer-aided design (CAD) models under various operating conditions.

[0044] Figure 2 yes Figure 1 Schematic diagram of a more detailed representation of the method shown.

[0045] Figure 2 The method represented in includes training a machine learning software model 106 using both real data about the process (in the form of machine process parameters 102a and real sensor data 102b) and virtual data 104 to be able to predict (at 214) the behavior of real sensors in a real version of the process.

[0046] The method includes providing machine process parameters 102a and / or model data (at 216) to facilitate simulation of the process (at 218) to generate virtual data 104. As an example, the model data can be a computer-aided design (CAD) model of a product to be manufactured by a 3D printing process. However, the model data can actually be any other set of information related to a real-world process that can be used to help simulate the process. Generally, if the process to be simulated will include a machine that operates on process parameters, those process parameters 102a can also be provided into the simulation (at 218). In some embodiments, the simulation (at 218) can be performed based solely on the machine process parameters.

[0047] The illustrated method also includes executing the real process (at 220). Sensors are provided to sense certain features of the process. In this regard, the sensors may be deployed inside a machine used to execute the process or certain aspects of the process. Thus, executing the process (at 220) generates more real data in the form of real sensor data 102b.

[0048] The real data (including machine process parameters 102a and real sensor data 102b) and virtual data 104 generated by the simulation (at 218) are fed into the machine learning software model 106 to train the machine learning software model 106 to be able to predict the behavior of the sensors.

[0049] Once trained, according to the illustrated embodiment, the machine learning software model 106 predicts real sensor values. The machine learning software model 106 can make these predictions based on any combination of real data 102 (different from real sensor data because real sensor data will not be available for a process that has not yet occurred) and / or virtual data 104. In this regard, the machine learning software model 106 behaves like a transfer function that models sensor behavior for any possible combination of model and machine or process inputs.

[0050] Of course, the techniques disclosed herein can be applied to a variety of possible applications. One such application is additive manufacturing, such as 3D printing. The nature of 3D printing is that there may not be enough time to conduct extensive training of the machine learning software model 106 or to simulate / model the precise physics of the 3D printing process, 3D printers, and / or photodiode sensors in the 3D process.

[0051] Figure 3 is a schematic cross-sectional view of an exemplary 3D printer 322 .

[0052] The illustrated 3D printer 322 is generally configured to generate functional engineering components (parts) one layer at a time from computer-aided design (CAD) data. In a typical embodiment, the 3D printer 322 produces a certain amount of material in a layered manner by melting material feedstock (e.g., powder, wire, etc.) using an energy source (e.g., laser, electron beam, arc, etc.) following a tool path derived from the CAD model.

[0053] The 3D printer 322 in the illustrated embodiment includes: a housing 324 defining a pair of internal cylinders; a build platform 326 movable in a vertical piston-like manner in a first of the cylinders relative to the housing 324; a new powder storage platform 328 that moves in a vertical piston-like manner in a second of the cylinders; and a powder roller 330. In a typical embodiment, during operation, the new powder storage platform 328 pushes the new powder storage 332 upward, the build platform 326 moves in a downward direction, and the powder roller 330 rolls the new powder from the new powder storage off the powder bed 331.

[0054] The 3D printer in the illustrated embodiment has a heat source 334, which can be a laser, an electron beam source, etc. During operation, the heat source 334 moves across the top of the powder bed to direct thermal energy (e.g., in the form of a laser beam) onto different portions of the powder bed to melt the upper layer of powder in each portion of the powder bed. The heat from the heat source melts the powder as it moves through the powder bed. After the laser passes through a specific point on the powder bed, the melted powder cools and solidifies into a solid (non-powder) form, thereby forming a portion of the part being printed. The 3D printer in the illustrated embodiment also has a photodiode sensor 336 near the laser delivery point on the powder bed. The photodiode sensor 336 is configured to sense light intensity (as an indicator of the thermal energy delivered to the powder bed by the heat source 334 at each point in time). More specifically, in a typical embodiment, the photodiode sensor includes a combined effect of energy delivered, material absorption rate, conduction, radiation, airflow, light reflection, etc.

[0055] Figure 4 yes Figure 2 Schematic diagram of an implementation of the method represented in a 3D printing environment. Although specific to the 3D printing environment, Figure 4 The method presented in can be easily applied to a variety of applications that may benefit from fast training and fast prediction.

[0056] Figure 4The method represented in includes training a machine learning software model 406 with both real data (in the form of machine process parameters 402a and photodiode sensor data 402b) and virtual data to be able to predict (at 414) the real behavior of the photodiode sensor 336. The virtual data in the illustrated embodiment includes physics-based features 404a, which represent the energy introduced into the part being printed within a specific time window, and the ability of the system to dissipate the introduced energy via thermal conduction. There are many possible ways to generate (at 403) these physics-based features. In a typical embodiment, the physics-based features are generated (at 403) by software executed on a computer-based processor and can be based on any one or more of a variety of inputs. These inputs can include, for example, machine process parameters 402a of a 3D printer and / or information from a high-resolution machine tool path 404b. The machine tool path 404b in the illustrated embodiment is generated by a powder bed manufacturing simulator 418 based on a CAD model 416 of the part to be manufactured.

[0057] therefore, Figure 4 The process represented in includes generating a CAD model 416 of certain geometric shapes (e.g., parts) to be printed by a 3D printer. Figure 5 Several examples of CAD models 501, 503, 505, 507, 509, 511, 513 and 515 with different geometries represented in the CAD models are shown in the figure above the virtual build platform 517 of the 3D printer.

[0058] Figure 4 The process shown in includes determining (at 402a) machine processing parameters for a 3D printer to build a part.

[0059] There are many ways to determine the machine processing parameters 402a. In some embodiments, the technician will look at the geometry of the part and use a set of machine parameters based on experience (e.g., uniform laser power for laser shadow line scanning, distance between shadow line scans, etc.). The technician can then enter the parameters into the real machine (i.e., real 3D printer) or virtual machine (i.e., virtual simulation of 3D printer / photodiode sensor) in the powder bed manufacturing 418 application. The real machine will generate machine controls to build the part accordingly, or the virtual machine will generate a machine tool path for virtual printing simulation. The quality of the real part coming out of the machine is usually checked, and if necessary, the machine settings can be adjusted based on the inspection, and then the part is printed again. This trial and error process may continue until an acceptable level of quality is reached. On the virtual side, engineers can check the virtual part and adjust the settings in the powder bed manufacturing application 418 as needed.

[0060] Figure 6 and Figure 7 A screenshot of a powder bed manufacturing application (418) is shown that enables a user to input various machine process parameters for a particular printing process.

[0061] More specifically, the screenshots shown enable the user to specify global parameters for the printing process (in Figure 6 in), composition strategy (in Figure 7 ) and identifying the scan path sequence (not shown in a separate figure).

[0062] Figure 6 The Global Parameters page in the screenshot enables the user to specify or enter a name for the scan path rule and identify the slice step size (in millimeters) for printing. The screenshot enables the user to specify upskin information including minimum skin width, number of layers, and the ability to limit the core by upskin. Similarly, the screenshot enables the user to specify downskin information including minimum skin width, number of layers, and the ability to limit the core by downskin.

[0063] exist Figure 7 The Composition Strategy page in the screenshot of FIG. 1 enables the user to specify the region type, contour information, fill information, definition parameters, and strategy information including XY strategy information. Contour information includes depth, definition, periodicity, start, range, and action information. Fill information includes depth, definition, periodicity, start, range, and action information. Definition parameters include scanning rules. Strategy information includes fill mode, angular rotation between slices, hatch direction, radial step size, and offset from the previous contour.

[0064] The Scan Path Sequence page typically includes a list of definitions created for scan path sequences (e.g., Outline, Shadow, etc.).

[0065] Reference again Figure 4 , based on the provided information (based on the CAD model and / or machine process parameters); the powder bed manufacturing application (at 418) in FIG. 4 generates a high-resolution machine tool path 404b. An example of this high-resolution machine tool path 404b information generated by the powder bed manufacturing application (at 418) is shown in FIG. Figure 8 and Fig. 9 Shown in. Figure 8 and Fig. 9 are exemplary screenshots generated by a powder bed manufacturing application 418 showing information about a detailed tool / machine path 404b generated by the application 418 for an array of cubic parts on a build platform. The arrows and lines in these screenshots represent the direction of travel and path of the laser head.

[0066] Figure 8 The screenshot in includes an image field 811 with a three-dimensional visual representation of the laser scan path used to print a part. Figure 8 The screenshot in also includes a first user menu 813 located on the right side of the screen, which provides the user with access to the Powder Bed Manufacturing Wizard functions, including functions related to Machine, Build Tray, Powder, Layout, Supports, Scan Path, Analysis, and Output. A second user menu 815 appears at the bottom of the screenshot and provides the user with access to the Powder Bed Manufacturing functions, including Standards, Setup, Programming, Analysis and Output, View, AR-VR, Tools, and Touch.

[0067] Fig. 9 The screenshot in has an image field 911 of a three-dimensional visual representation showing a portion of a laser scan path used to print a part. Fig. 9 The screenshot in also includes a user menu 913 located on the left side of the screen, which provides the user with access to various powder bed manufacturing functions related to displaying slice and scan paths, trajectories, and types of scan paths.

[0068] Typically, information accessible from the high-resolution machine tool path 414 (e.g., in a powder bed manufacturing application environment) includes information related to the machine tool path of the part represented by the CAD model 416. This information typically includes laser power information, laser speed information, and laser pattern information. In an exemplary embodiment, the high-resolution tool path contains temporal and spatial information of the laser path, pattern, and state. In this regard, a tool path segment may be as follows:

[0069] <Time 1>, <x1> , <y1> , <z1>, <Power 1>

[0070] <Time 2>, <x2> , <y2> , <z2>, <Power 2>

[0071] <Moment 3>, <x3> , <y3> , <z3>,<Power 2>

[0072] The tool path segment information indicates that the laser should travel from point 1 at time 1, x1, y1, z1 coordinates to point 2 at time 2, x2, y2, z2 coordinates at a constant speed equal to the distance between point 1 and point 2 divided by the difference between time 1 and time 2, and a constant power equal to power 1. The laser then travels from point 2 at time 2, x2, y2, z2 coordinates to point 3 at time 3, x3, y3, z3 coordinates at a constant speed equal to the distance between point 2 and point 3 divided by the difference between time 2 and time 3, and a constant power equal to power 2, and so on.

[0073] The illustrated method also includes performing one or more real-world printings of the part (at 420), which results in real-world photodiode sensor data 402b. Thus, the photodiode sensor data 402b is generated by one or more real machines during one or more real printing processes. In some embodiments, to begin the machine learning process, some parts are programmed using various machine process parameters 402a and first printed to generate an initial set of photodiode sensor data 402b for training purposes. The data format will depend on the specific machine and sensor, but in a typical embodiment, the data format will include at least time, spatial location, and sensor amplitude information. The amount of data typically depends on the part size, machine parameters, and sensor output frequency.

[0074] In the illustrated embodiment, the physics-based features 403 are calculated based on information from a high-resolution machine tool path 404b.

[0075] As described above, the high-resolution machine tool path 404b specifies the laser position at different times, the laser power at these different times, and the laser speed between these times. In addition, in a typical embodiment, a region of interest is specified around the laser position (specified by the user or as a process preset) and stored in a computer-based memory. The size and shape of the region of interest can vary depending on the application. However, in a typical embodiment, the region of interest will be a circular area on the upper surface of the powder bed / part being printed, which has the same (+ / -10%) dimensions as the printed portion of the powder bed / part that the photodiode sensor is configured to sense. Therefore, if the photodiode sensor is configured to sense a circular area with a diameter of 4 mm, the region of interest can also be a circle with a diameter of 4 mm (+ / -10%). The region of interest moves with the laser position on the surface of the powder bed or part being printed at each point during the printing process and generally concentrically surrounds the laser position on the surface of the powder bed or part being printed.

[0076] For example, based on the above information, a computer-based processor can determine how much energy to deliver by the laser to the upper surface of the powder bed and part being printed during the printing process, and exactly where to deliver that energy to any location and for any period of time.

[0077] In a typical embodiment, a computer-based processor calculates an energy flux signature that represents how much energy has been delivered by the laser to a region of interest surrounding a laser position during a specific time period. The time period may include any portion or all of a printing process and may have been defined by a user or set by default. The computer-based processor may calculate the energy flux signature by integrating the laser power / energy data delivered to the region of interest during the specific time period. The computer-based processor may update its energy flux signature calculation periodically or nearly continuously and / or in near real time throughout the printing process.

[0078] In a typical embodiment, the computer-based processor also calculates an effective conduction characteristic that provides a measure of the ability of the powder bed and / or part to dissipate energy delivered by the laser at a particular point in time. In an exemplary embodiment, the computer-based processor calculates the effective conduction characteristic by calculating the area (e.g., in square millimeters) within the region of interest that has not been scanned by the laser for a particular period of time. The computer-based processor can update its effective conduction characteristic calculations to coincide with updates in its energy flux characteristic calculations. Thus, they can be updated periodically throughout the printing process, or updated nearly continuously and / or in nearly real time.

[0079] For example, by referring to Figures 10 to 12 The concepts of energy flux and efficient conduction in an environment can be further understood in physics-based characteristic calculations.

[0080] Fig.10 To illustrate the use of a 3D printer (e.g., Figure 3 A top view of a powder bed 1351 in which a part is being printed in a 3D printer 322 in the drawing, and a schematic diagram of the path that a laser tool will travel along the upper surface of the powder bed in a rectangular print layer.

[0081] The outer rectangle 1350 shown as a dotted line represents the boundary of the current layer part. The two solid rectangles 1352 within the dotted line represent the laser paths of the two contour lines. The solid and dotted lines within the contour line boundary represent the path of the fill laser tool. Specifically, the solid line 1354 within the contour line boundary represents the area within the area that the laser has scanned as of a specific point in time, and the dotted line 1356 within the contour line boundary represents the area within the area that the laser has not scanned as of a specific point in time. Fig.10 The point 1358 near the center of represents the laser position at a particular point in time during the scanning process of the laser. The circle 1360 surrounding the black dot represents the region of interest used to calculate the physics-based features 404a (e.g., energy flux features and effective conduction features). The region of interest is essentially the area within the region of the powder bed / part being printed that is considered relevant for the purpose of calculating the above-mentioned physics-based features.

[0082] In a typical embodiment, the boundary (outer dotted line 1350) at which the part is located in the current layer can be based on the size of the part being printed and / or the size of the powder bed 328. The two contour laser paths (solid rectangular line 1352) and the path of the fill laser tool (solid line 1354 and dashed line 1356 within the contour boundary) are functions of the high-resolution machine tool path 404b. In some embodiments, the size and shape of the area of ​​interest 1360 is the same as the area on the upper surface of the powder bed 328 that the photodiode sensor 336 is configured to sense (+ / -10%). In an exemplary embodiment, the photodiode sensor 336 is configured to sense the area within a circular area with a diameter of approximately 4 mm on the upper surface of the powder bed 328 / part being printed. In those embodiments, the diameter of the circle identifying the area of ​​interest can be set to approximately 4 mm (+ / -10%). In some embodiments, the size of the circle may change depending on one or more factors (e.g., the speed of the laser 334 passing through the powder bed 328, etc.). In some implementations, the size of the circle may also be specified by the user (e.g., at a computer-based interface that enables the user to interact with the physics-based feature generation process). In a typical embodiment, as the laser moves across the powder bed surface, the area of ​​interest circle will also move across the powder bed surface, such that the laser position point 1358 moving across the powder bed surface is always centered within the area of ​​interest circle 1360.

[0083] The thicker solid lines 1362 within the area of ​​interest circle 1360 represent the areas within the area of ​​interest circle 1360 that have been scanned by the laser during a particular period of time. Moreover, these thicker solid lines 1362 identify the areas within the area of ​​interest circle 1360 that are considered relevant for the purpose of calculating the energy flux characteristics corresponding to the laser position shown by point 1358.

[0084] The narrower solid lines 1364 within the area of ​​interest circle 1360 identify areas within the region that have been scanned by the laser, but are not relevant for purposes of calculating the energy flux signature for the indicated laser position 1358. One reason why the areas within the regions identified by the narrower solid lines are not considered relevant in this regard may be that those scans occurred outside of the particular time period that is considered relevant for energy flux calculations. Essentially, if enough time has passed since a particular area was laser scanned, the energy dissipated in that area will be large enough that the area within that area can be ignored when calculating the energy flux signature.

[0085] Dashed lines 1366 within the area of ​​interest circle 1360 identify areas within regions that have not yet been scanned by the laser. These areas (identified by dashed lines 1366) are considered irrelevant for the purpose of calculating the energy flux characteristics of the indicated laser position 1358. They are considered irrelevant for the purpose of calculating the energy flux characteristics because laser energy has not yet been delivered to these areas. However, the areas within these regions are considered relevant for calculating the effective conduction characteristics. And further, the areas within these regions are considered relevant for calculating the effective conduction characteristics because these regions have not yet received direct delivery of laser energy. As such, these regions 1366 are much cooler than the most recently scanned regions 1362, and therefore, these regions 1366 provide an escape path for heat / energy from region 1362 by thermal conduction. In some embodiments, region 1364 may also be included in the calculation of effective conduction because, if sufficiently cooled, this region 1364 may also provide an escape path for heat / energy from region 1362 by thermal conduction. Generally, if the area 1366 (and the area 1364) is larger, a stronger heat dissipation capability will be produced; generally, if the area 1366 (and the area 1364) is smaller, a weaker heat dissipation capability will be produced.

[0086] The computer-based processor may calculate the effective conductive feature in a variety of ways. In one such example, the computer-based processor may subtract the area of ​​region 1362 from the area of ​​the entire region of interest 1360. In another such example, the computer-based processor may calculate the area of ​​region 1366. In yet another such example, the computer-based processor may add the area of ​​region 1366 to the area of ​​region 1364. In some embodiments, these calculations may be performed directly from information provided by high-resolution machine tool path 404b.

[0087] Fig.11 and Fig.12 for Fig.10 Schematic diagram of the powder bed 1351 in FIG. 1 , which shows that at the beginning of the scanning process ( Fig.11 during scanning) and when the scanning process is about to end ( Fig.12 Laser position 1358 and area of ​​interest circle 1360 near the center).

[0088] These figures help illustrate several points. First, the area of ​​interest circle 1360 moves with the laser position 1358 throughout the printing process, keeping the circle 1360 centered. Second, the calculations for physics-based features discussed above are applicable and accurate even when the laser is close to the geometric boundaries of the part being created. Fig.11 The bold line in represents the integration path used to calculate the unscanned area, which is a relatively large portion of the region of interest (ie, a characteristic size local patch). Fig.12 The bold line in represents the integration path used to calculate the unscanned area, which is a relatively small portion of the circle. Therefore, the physics-based features (i.e., energy flux features and effective conduction features) will automatically incorporate geometric boundary effects, eliminating the need to engineer geometric features into the machine learning process, which can be extremely complex and difficult to meet.

[0089] Various other techniques or methods for calculating the physics-based features are possible. However, in general, these physics-based features 404a can be quickly and easily calculated even without extensive simulation or modeling. Moreover, these features have a significant impact on enabling the machine learning software model 406 to accurately predict the behavior of the photodiode sensor 406b. Furthermore, it has been discovered that including the physics-based features 404a in the training process of the machine learning software model 406 enables the machine learning software model 406 to better predict the behavior of the photodiode sensor in other processes / machines. Additionally, by accessing detailed tool path information (e.g., according to Dassault Systèmes SE These methods can be easily implemented and adopted using virtual machine data provided by software platforms, and due to their relative simplicity and efficiency, they can often be extended to part-level predictions.

[0090] Reference again Figure 4 , physics-based features 404a (i.e., one or more energy flux features and one or more effective conduction features) are provided to a machine learning software model 406 along with machine process parameters 402a and real-world photodiode sensor data 402b to train the machine learning software model 406 to be able to predict photodiode sensor values ​​414.

[0091] In some embodiments, particularly those where speed of training and / or prediction is of particular concern, the aforementioned physics-based features (i.e., energy flux features and effective conduction features) may be the only form of virtual data used to train the machine learning software model 406. In those cases, all remaining training data provided to the machine learning software model 406 may be real data (e.g., machine process parameters 402a of a 3D printer and / or real photodiode sensor data 402b). In some embodiments, particularly where there are not strict constraints on available time, it may be desirable to supplement training with other virtual data.

[0092] It has been discovered that the phenomena represented by these specific physics-based features (i.e., the energy flux feature and the effective conduction feature) can be generated very quickly and are very effective in training the machine learning software model 406 to accurately predict the behavior of the photodiode sensor 336 under a variety of operating conditions. Furthermore, high accuracy can be achieved even when these features are the only types of fictitious data used in the training process, and all other training data is real data.

[0093] Once trained, according to the illustrated embodiment, the machine learning software model 406 is able to predict (at 414) the behavior of a real photodiode sensor.

[0094] It has been found that machine learning software models trained using physics-based features (in addition to real data) can produce more accurate sensor predictions than machine learning software models trained using only real data. Fig.13 shows a single layer of test data prediction scatter plot based only on machine process parameters and photodiode sensor data, and Fig.14 Predicted scatter plots for a single layer of test data based on machine process parameters, photodiode sensor data, physics-based features, and CAD models are shown. These plots show a significant improvement in accuracy when physics-based features and CAD models are included.

[0095] The trained machine learning software model can then be used to provide a quick prediction of the physical sensor data for any given part to guide the machine operator to a better machine setting; or to optimize the machine parameters of the target sensor signal for the entire part, thereby reducing trial and error and achieving the right printing effect. For example, based on the initial setting of the machine process parameters for a given CAD geometry, the above method can be used to generate physics-based features and combined with real data (e.g., initial machine process parameters) to train the machine learning software model to predict the sensor readings for the entire part. Then, the optimal machine process parameters, including the optimal 3D laser power or laser velocity field for the entire part, can be repeatedly derived using, for example, a sensitivity-based optimization method.

[0096] Fig.15 An experimental validation of the photodiode sensor readings for a single layer before and after a print optimization process relying on this approach is shown. The old photodiode sensor readings with the initial machine process parameters and the new photodiode sensor readings with the optimal machine process parameters are both labeled. The results show that through print optimization, the photodiode sensor can be optimized very close to the target level, thereby preventing the part from overheating or underheating during the printing process and significantly reducing the trial and error required to successfully build.

[0097] Fig.16 A schematic diagram of an exemplary process of a machine learning and printing optimization framework is shown. In the illustrated embodiment, a training data set is first defined (at 1970). Physical experiments are performed to record machine settings and sensor data. Using the proposed method of these parts, relevant physics-based features are created (at 1972). The data is then used to train a predictive machine learning software model. Given a new part CAD model (at 1974) and machine operating constraints, the machine settings are then optimized for target sensor output (at 1976). A machine-readable file format is generated and sent to the hardware (at 1978) to print with the target part quality for the first time. As more parts are printed and data is added as additional training data, the machine learning software model can continue to be trained, and as more parts are printed, the accuracy of the machine learning model will also increase.

[0098] Once the machine learning software model is trained, the trained machine learning software model, physics-based feature generation, and optimization methods can be used in conjunction with one or more additive manufacturing machines to make sensor predictions and optimize machine parameters in real time on a layer-by-layer basis. Fig.17 An exemplary implementation of such a process is shown.

[0099] Fig.17 The process shown includes importing a CAD model into the machine and setting initial machine parameters (at 2080) and deploying optimization for the first layer (at 2082). Typically, optimization for the first printed layer is done based on machine learning predictions using the machine parameters and the generated physics-based features. Then (at 2084), the machine prints the first layer using the optimized parameters. While the first layer is being built, the system deploys optimization for the second layer (at 2086). Then (at 2088), the machine prints the second layer using the optimized parameters. While the second layer is being built, the system deploys optimization for the third layer (at 2090). This process continues until the last layer is built and until the last layer is built (at 2092).

[0100] Fig.18 In many ways it is similar to Fig.16 The schematic diagram shown. Fig.18 In the example of the embodiment of the present invention, the sensor and operating condition data are first defined (at 2170). Physical experiments are conducted to record machine settings and sensor data. Simulation data is created (at 2172). The predictive machine learning software model is then trained using this data. The machine learning software model is then used to make sensor predictions in service (at 2174). Next, the system determines (at 2176) whether there is a sensor anomaly (i.e., whether the actual sensor reading differs from the predicted sensor reading by a sufficient amount). In a typical embodiment, the system (e.g., a computer-based processor in the system) can receive the actual sensor reading and compare the actual sensor reading with the corresponding predicted sensor reading, and then, if the difference between the two exceeds a certain threshold (e.g., 1%, 5%, 10%, etc.), a warning is issued (at 2178). In some embodiments, the threshold can be defined by the user. In some embodiments, the threshold can be a preset threshold. The warning can take any of a variety of different forms, including, for example, a message sent to an operator, which can have an audio-visual element or a tactile element, etc. In some embodiments, the warning can be accompanied by a pause in the printing process. And other warnings and associated operations are also possible.

[0101] Fig.19 A computer network or similar digital processing environment is shown in which embodiments of the present invention may be implemented.

[0102] The client computer / device 50 and the server computer 60 provide processing, storage, and input / output devices for executing application programs, etc. The client computer / device 50 may also be connected to other computing devices, including other client devices / processes 50 and server computers 60, via a communication network 70. The communication network 70 may be part of a remote access network, a global network (e.g., the Internet), a cloud computing server or service, a worldwide collection of computers, a local area network or a wide area network, and gateways that currently use various protocols (TCP / IP, Bluetooth, etc.) to communicate with each other. Other electronic device / computer network architectures are also applicable.

[0103] Fig. 20 for Figure 3 80. The computer system of FIG. 80 is a block diagram of the internal structure of a computer (e.g., a client processor / device 50 or a server computer 60) in FIG. 80. Each computer 50, 60 includes a system bus 79, where a bus is a set of hardware lines used to transfer data between components of a computer or processing system. The bus 79 is essentially a shared channel that connects the different elements of the computer system (e.g., processor, disk storage, memory, input / output ports, network ports, etc.), which enables information to be passed between the elements. An I / O device interface 82 is attached to the system bus 79 for connecting various input and output devices (e.g., keyboard, mouse, display, printer, speakers, etc.) to the computer 50, 60. A network interface 86 enables the computer to connect to a network (e.g., Figure 3 The memory 90 is used to implement computer software instructions 92 and data 94 (for example, the above Figure 1 and 2 84, or a combination of CPUs 84 ...

[0104] In one embodiment, the processor routine 92 and data 94 are computer program products (generally referred to as 92), which include computer readable media (such as one or more high-density read-only optical disks (Digital Video Disk-Read Only Memory, DVD-ROM), read-only optical disk drives (CompactDisc Read-Only Memory, CD-ROM), disks, tapes, and other removable storage media) that provide at least part of the software instructions for the system of the present invention. The computer program product 92 can be installed by any suitable software installation process, which is well known in the art. In another embodiment, at least part of the software instructions can also be downloaded through a cable, communication and / or wireless connection. In other embodiments, the program of the present invention is a computer program propagation signal product 107, which is specifically a propagation signal on a propagation medium (for example, radio waves, infrared rays, laser waves, sound waves, or radio waves propagated on a global network such as the Internet or other networks). This carrier medium or signal provides at least part of the software instructions for the routine / program 92 of the present invention. In an optional embodiment, the propagated signal is an emulated carrier or digital signal carried on a propagation medium. For example, the signal propagated can be a digitized signal propagated on a global network (e.g., the Internet), a telecommunications network, or other network. In one embodiment, the signal propagated is a signal transmitted through a propagation medium over a period of time, such as instructions for a software application program sent in a packet form over a network in milliseconds, seconds, minutes, or longer. In another embodiment, the computer-readable medium of the computer program product 92 is a propagation medium that can be received and read by the computer system 50, such as by receiving the propagation medium and identifying the propagation signal contained in the propagation medium for the computer program propagation signal product as described above. In general, the term "carrier medium" or transient carrier includes the aforementioned transient signal, propagation signal, propagation medium, storage medium, etc. In other embodiments, the program product 92 can be implemented as a so-called software as a service (SaaS) or other installation or communication that supports end users.

[0105] Compared with the use of experimental data alone, virtual simulation data can provide more physical information about the system. For example, simulation can tell not only the temperature at the sensor installation location, but also the temperature at various locations in the system; the sensor is limited by the operation of the device, such as the measurement range or the operating parts, but the simulation data is always accessible; simulation provides additional physical quantities, such as thermal radiation, stress, strain, etc., which cannot be actually measured by experiment, but are crucial for predicting system results. The present invention proposes a machine learning service based on virtual simulation and real (Virtual+Real, V+R) data. Simulation data includes high-fidelity physical simulation models (including but not limited to finite element modeling) and data, as well as reduced-order or first-order models and data developed from feature importance and intelligence obtained or verified from the machine learning process. Real data includes: real parameters such as design parameters, geometric structures, materials, manufacturing settings, etc., which are all known without knowing modeling and simulation.

[0106] A number of embodiments of the present invention have been described. However, it should be understood that various modifications can be made to the described embodiments without departing from the spirit and scope of the present invention.

[0107] For example, the disclosed systems and methods can be applied to a wide range of processes including all types of manufacturing processes (including additive manufacturing processes and other non-manufacturing processes). Once trained, machine learning software models can be applied to a variety of different purposes in the process of interest, such as prediction, diagnosis, simulation, etc.

[0108] The specific combination of real data and virtual data used to train machine learning software can vary widely. In most cases, the training method can be tailored to the practical considerations involved in a particular process. For example, if time is of the essence, the physics-based features disclosed herein may be the only (or nearly the only) virtual data used in the training process. However, other forms of virtual data may be used when more time is available. Also, various types of real data may be used.

[0109] Almost any type of computer-aided drawing application can be used to generate a CAD model, and the CAD model can be very detailed or not very detailed. Similarly, any one or a combination of multiple different simulation applications can be used to generate simulation. A machine process can be a process comprising one or more machines, and the one or more machines can work in a serial, parallel or combination thereof. The sensor to be predicted can be any type of sensor, for example, for optics, light, imaging, photons, acoustics, sound, vibration, automobile, chemistry, electric current, electric potential, magnetism, radio, environment, weather, humidity, flow, fluid velocity, ionizing radiation, subatomic particles, navigation, position, angle, displacement, distance, speed, acceleration, pressure, force, density, level, heat, heat, temperature, proximity, existence, speed, etc.

[0110] The machine learning software model may be any type of machine learning software model that is trainable and executable in a computer-based environment.

[0111] The optimization process disclosed herein can include any type of process that is iteratively performed by comparing various solutions until an optimal or satisfactory solution is found. The trained machine learning software model facilitates the optimization because it feeds instant sensor predictions back to the optimizer at each iteration, thereby reducing the overall time required to achieve a particular degree of optimization goal.

[0112] The 3D printer disclosed herein is only one example of a 3D printing machine. Many variations are possible. The sensor discussed herein with particular reference to 3D printers is a photodiode sensor. However, other sensors are also possible.

[0113] This application discloses a powder bed manufacturing application as an example of an application that can generate high resolution machine tool paths. However, other tools may also be able to generate machine tool paths.

[0114] The physics-based features that the application focuses on include energy flux features and effective conduction features. In various embodiments, other physics-based features can also be calculated and used to train the machine learning software model.

[0115] Furthermore, although this specification contains details of many specific embodiments, these should not be construed as limitations on any invention or the scope of what may be claimed, but rather merely as descriptions of particular features of particular embodiments of particular inventions. Certain features described in the context of a single embodiment of this specification may also be implemented in a combination that includes that single embodiment. Conversely, various features described in the context of a single embodiment of this specification may also be implemented in multiple embodiments, respectively, or in any suitable subcombination thereof. Moreover, although the feature descriptions hereinabove may be described as functioning in certain combinations, and may even be initially claimed to be so, in some cases, one or more features of a claimed combination may also be excised from the combination, and a claimed combination may be directed to a subcombination or a variation of a subcombination.

[0116] Similarly, while operations disclosed herein occur in a particular order, this should not be understood as requiring that such operations be performed in the particular order shown or in sequential order, or that all illustrated operations be performed, to achieve desirable results.

[0117] Other implementations are within the protection scope of the claims of this application. < / y3> < / x3> < / y2> < / x2> < / y1> < / x1>

Claims

1. A method comprising: collecting and using process parameters from a real-world additive manufacturing machine, real-world sensor readings from said additive manufacturing machine manufacturing a real-world part, and physics-based features generated by a computer-based processor to train a machine learning software model to predict future behavior of said real-world sensors in a subsequent real-world version of an additive manufacturing process by said real-world additive manufacturing machine; The collection includes: inputting machine process parameters of said additive manufacturing process into a real-world additive manufacturing machine to produce a part via an additive manufacturing process; providing a real-world sensor to sense a characteristic associated with said additive manufacturing process; manufacturing a first real-world part using the real-world additive manufacturing machine through the additive manufacturing process; When the real-world machine manufactures the first real-world part through the additive manufacturing process, sensing a characteristic associated with the additive manufacturing process using the real-world sensor; receiving a sensor reading from the real-world sensor while the real-world machine is manufacturing the real-world part via the additive manufacturing process; generating, with the computer-based processor, a physics-based signature associated with the additive manufacturing process; and Wherein, the use includes: A machine learning software model is trained based on the machine process parameters, the real-world sensor readings, and the physics-based features to predict behavior of the real-world sensor.

2. The method according to claim 1, further comprising: Providing a computer-aided design (CAD) model of the part; as well as generating virtual data about a machine tool path based on the CAD model using computer-based software, the virtual data about the machine tool path being used to generate the part in the additive manufacturing process, Wherein generating the physics-based features is based at least in part on virtual data regarding a machine tool path.

3. The method according to claim 2, wherein: The physics-based features represent energy delivered to and heat conducted away from a region of interest during the additive manufacturing process.

4. The method according to claim 3, wherein: The region of interest is a region of the surface of the powder bed, wherein the region of interest is concentric with the spot at which energy is introduced into the powder bed at a specific point in time during the additive manufacturing process.

5. The method according to claim 4, wherein: As the additive manufacturing process progresses, the region of interest moves and remains concentric with the location where energy is introduced into the powder bed.

6. The method according to claim 2, wherein: Generate physics-based features including: Generate energy flux features, The energy flux characteristic corresponds to the amount of energy delivered to the region of interest within a specific time period during the additive manufacturing process.

7. The method according to claim 6, wherein: The generated energy flux features include: The computer-based processor is employed to integrate power entering the region of interest over a specified period of time based at least in part on virtual data regarding a machine tool path.

8. The method according to claim 7, wherein: Virtual data about the machine tool path includes time, location, and power information of heat sources in the additive manufacturing machine.

9. The method according to claim 2, wherein: Generate physics-based features including: Generate effective conduction characteristics, The effective conduction characteristic corresponds to the ability to dissipate energy introduced into the region of interest through thermal conduction.

10. The method according to claim 9, wherein: Generating effective conduction features includes: Based at least in part on the virtual data regarding the machine tool path, it is determined how much area of ​​the region of interest has not received direct delivery of energy within a limited time window.

11. The method according to claim 10, wherein: Virtual data about the machine tool path includes time, location, and power information of heat sources in the additive manufacturing machine.

12. The method according to claim 1, wherein: The machine process parameters include: power, speed, and scanning pattern information associated with the additive manufacturing process.

13. The method according to claim 1, further comprising: The process parameters are optimized using a trained version of the machine learning software application.

14. The method according to claim 13, wherein: Using the trained version of the machine learning software application to optimize the process parameters includes: The machine process parameters are adjusted during the additive manufacturing process based on information from the machine learning software model to generate desired sensor readings at the real-world sensor.

15. The method according to claim 1, further comprising: employing the machine learning software model to predict behavior of the real-world sensor during the additive manufacturing process; comparing the predicted behavior to actual behavior of the real-world sensor during the additive manufacturing process; as well as Anomalies in the behavior of the real-world sensor are detected based on the comparison.

16. The method according to claim 1, wherein: The additive manufacturing process is 3D printing, and wherein the additive manufacturing machine comprises: a build platform; a powder bed located above the build platform; and a heat source configured to deliver heat to a surface of the powder bed to generate the part.

17. A system for collecting and using process parameters from a real-world additive manufacturing machine, real-world sensor readings from the additive manufacturing machine manufacturing a real-world part, and physics-based features generated by a computer-based processor to train a machine learning software model to predict future behavior of the real-world sensors in a subsequent real-world version of an additive manufacturing process by the real-world additive manufacturing machine, the system comprising: A computer, the computer comprising: Computer-based processors; and a computer-based memory storing computer-executable instructions that, when executed by a computer-based processor, cause the computer-based processor to generate physics-based features associated with an additive manufacturing process; one or more real-world additive manufacturing machines configured to manufacture a real-world part by executing a real-world version of the additive manufacturing process based on machine process parameters input to the one or more machines and associated with the additive manufacturing process; a real-world sensor to sense a characteristic associated with a real-world version of the additive manufacturing process; and a machine learning software model trained to predict behavior of the real-world sensors based on physics-based features, the machine process parameters, and sensor readings from the real-world sensors sensed as the one or more real-world additive manufacturing machines manufactured the real-world part in a real-world version of the additive manufacturing process.

18. The system of claim 17, further comprising: Computer-based software for providing a computer-aided CAD model of said part; as well as computer-based software for generating virtual data regarding a machine tool path based on the CAD model, the virtual data regarding the machine tool path being used to generate the part in the additive manufacturing process, Wherein the physics-based features are based at least in part on virtual data regarding a machine tool path.

19. The system of claim 18, wherein: The physics-based features represent energy delivered to and heat conducted away from a region of interest during the additive manufacturing process.

20. The system of claim 19, wherein: The additive manufacturing machine comprises: a powder bed, and a heating device for delivering heat to the powder bed, wherein the region of interest is a region on the surface of the powder bed, and Therein, the region of interest is concentric with the spot at which energy is introduced into the powder bed by the heating device at a specific point in time during the additive manufacturing process.

21. The system of claim 20, wherein: As the additive manufacturing process proceeds, the region of interest moves and remains concentric with the location where energy is introduced into the powder bed by the heating device.

22. The system of claim 18, wherein: Generate physics-based features including: Generate energy flux features, The energy flux characteristic corresponds to the amount of energy delivered to the region of interest within a specific time period during the additive manufacturing process.

23. The system of claim 22, wherein: The generated energy flux features include: The computer-based processor is employed to integrate power entering the region of interest over a specified period of time based at least in part on virtual data regarding a machine tool path.

24. The system of claim 23, wherein: The additive manufacturing machine includes a heat source for delivering heat to the powder bed, and The virtual data about the machine tool path includes: time, position, and power information of the heat source related to the additive manufacturing process.

25. The system of claim 18, wherein: Generate physics-based features including: Generate effective conduction characteristics, The effective conduction characteristic corresponds to the ability to dissipate energy introduced into the region of interest through thermal conduction.

26. The system of claim 25, wherein: Generating effective conduction features includes: Based at least in part on the virtual data regarding the machine tool path, it is determined how much area of ​​the region of interest has not received direct delivery of energy from a heat source in the additive manufacturing machine within a limited time window.

27. The system according to claim 26, wherein: Virtual data about the machine tool path includes time, location, and power information of heat sources in the additive manufacturing machine.

28. The system of claim 1, wherein: The machine process parameters include: power, speed, and scanning pattern information associated with the additive manufacturing process.

29. The system of claim 1, wherein: The additive manufacturing process is 3D printing, and wherein the additive manufacturing machine comprises: a build platform; a powder bed located above the build platform; and a heat source configured to deliver heat to a surface of the powder bed to generate the part.

30. A method of predicting the behavior of a real-world sensor in a machine for performing an additive manufacturing process to produce a real-world part, the method comprising: inputting machine process parameters of said additive manufacturing process into a real-world additive manufacturing machine to produce a part via an additive manufacturing process; providing a real-world sensor to sense a characteristic associated with said additive manufacturing process; manufacturing a first real-world part using the real-world additive manufacturing machine through the additive manufacturing process; When the real-world machine manufactures the first real-world part through the additive manufacturing process, sensing a characteristic associated with the additive manufacturing process using the real-world sensor; receiving a sensor reading from the real-world sensor while the real-world machine is manufacturing the first real-world part via the additive manufacturing process, the sensor reading being indicative of the characteristic; generating, using a computer-based processor, physics-based features associated with the additive manufacturing process; training a machine learning software model based on the machine process parameters, the real-world sensor readings, and the physics-based features to predict behavior of the real-world sensor; as well as After the training, the machine learning software model is utilized to predict future behavior of the real-world sensor in a subsequent real-world version of the additive manufacturing process employing the real-world machine to produce one or more subsequent real-world parts.

Images