System and method for classifying, reporting, and adjusting behavior changes of additive manufacturing fleet

Abstract

Apparatuses and related methods for classifying and adjusting constructs across additive manufacturing machines are disclosed. An example apparatus includes learner circuitry to: process first data from a set of first constructs to learn a behavior; classifying each construction as a standard construction or a non-standard construction; modeling the learned behavior to form a standard reference behavior and a non-standard reference behavior, the standard reference behavior including a first feature, and the non-standard reference behavior including a second feature; and outputting the standard reference behavior and the non-standard reference behavior to classify the additional constructs. The apparatus comprises an evaluator circuit configured to: ingest second data of a second construction; comparing the second data with a standard reference behavior and a non-standard reference behavior; classifying the second construct as a standard construct or a non-standard construct; and, when the second build is classified as a non-standard build, outputting a corrective action.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. Provisional Patent Application No. 63 / 596,486, filed on November 6, 2023. U.S. Provisional Patent Application No. 63 / 596,486 is incorporated herein by reference in its entirety for all purposes. Technical Field

[0003] The present disclosure relates to systems and methods for monitoring additive manufacturing devices, and more particularly, to systems and methods for monitoring, analyzing, and adjusting build-level and cluster-level additive manufacturing machines and associated processes. Background Art

[0004] In additive manufacturing processes, such as melting layers of powder to make an object, there are several challenges in diagnosing aborted or failed builds, or identifying performance issues with an additive manufacturing device. Specifically, an expert must manually diagnose the build or device, which requires a lot of time and manpower. In addition, finding the root cause of a failure of an additive manufacturing device is a difficult and time-consuming process, which is more or less impossible during a build. Reliability and repeatability are customer expectations even between builds. Failure to meet this expectation renders a single additive manufacturing machine, as well as a fleet of such additive manufacturing machines, ineffective at best and useless at worst. Therefore, there is an unmet need to improve machine and build reliability, as well as build repeatability, in a fleet of additive manufacturing machines. BRIEF DESCRIPTION OF THE DRAWINGS

[0005] Figure 1 An additive manufacturing system of the present disclosure is depicted in accordance with one or more examples shown and described herein.

[0006] Figure 2 is a block diagram of an example system according to one or more examples shown and described herein.

[0007] Figure 3 Depicted are various internal components of a control component of an additive manufacturing system according to one or more examples shown and described herein.

[0008] Figure 4 Depicted are various internal components of a user computing device in communication with an additive manufacturing system according to one or more examples shown and described herein.

[0009] Figure 5 An example infrastructure or framework for monitoring and adjusting additive manufacturing based on machine health monitoring and analysis is shown.

[0010] Figure 6 An example additive manufacturing machine behavior device or infrastructure is shown.

[0011] Figure 7 Shows Figure 6 Additional example views of the Additive Manufacturing machine behavior device.

[0012] Figure 8 Shows Figure 6 Additional example views of the Additive Manufacturing machine behavior device.

[0013] Fig. 9 Depicted are example output populations for behaviorally labeled builds.

[0014] Figure 10-12 It is used for implementation Figure 6-8 A flowchart of example hardware logic, machine readable instructions, hardware implemented state machines, and / or any combination thereof of an example additive manufacturing machine behavior device.

[0015] Figure 13-15 is structured to execute Figure 10-12 Instructions to implement Figure 6-8 Block diagram of an example processor platform and associated circuitry for an example additive manufacturing machine behavior device.

[0016] Fig.16 is used to transfer software, instructions and / or firmware (e.g., corresponding to Figure 10-12 A block diagram of an example software / firmware / instruction distribution platform (e.g., one or more servers) for distributing example machine-readable instructions of the present invention to client devices associated with end users and / or consumers (e.g., for licensing, sale and / or use), retailers (e.g., for sale, resale, licensing and / or sub-licensing), and / or original equipment manufacturers (OEMs) (e.g., for inclusion in products to be distributed to, for example, retailers and / or other end users (such as direct purchasing customers)).

[0017] The drawings are not drawn to scale. On the contrary, the thickness of the layer or region can be magnified in the drawings. Generally, the same reference numerals will be used throughout the drawings and the accompanying written description to refer to the same or similar parts. As used in this patent, it is stated that any part (e.g., layer, film, zone, region or plate) is in any way (e.g., positioned, located, set or formed, etc.) on another part, indicating that the referenced part is in contact with another part, or the referenced part is above another part, with one or more intermediate parts located between them. Unless otherwise stated, connection references (e.g., attachment, connection, connection and engagement) should be interpreted broadly and may include intermediate members between element sets and relative movement between elements. Therefore, connection references do not necessarily infer that two elements are directly connected and have a fixed relationship to each other. Declaring that any part is "in contact" with another part means that there is no intermediate part between the two parts.

[0018] Descriptors "first," "second," "third," etc., are used herein when identifying multiple elements or components that can be referred to separately. Unless otherwise specified or understood based on the context of their use, such descriptors are not intended to confer any meaning of priority, physical order or arrangement in a list, or temporal ordering, but are merely used as labels to refer to multiple elements or components separately to facilitate understanding of the disclosed examples. In some examples, the descriptor "first" may be used to refer to an element in the detailed description, while a different descriptor (e.g., "second" or "third") may be used in the claims to refer to the same element. In this case, it should be understood that the use of such descriptors is only for the convenience of referencing multiple elements or components. DETAILED DESCRIPTION

[0019] Additive manufacturing is the process of building a three-dimensional structure, typically in a series of layers, based on a digital model of the structure. While some examples of additive manufacturing techniques rely on sintering or melting / fusing using an energy source to form a structure, rather than "printing" which deposits material in selected locations, the term "printing" is often used to describe additive manufacturing processes (e.g., three-dimensional (3D) printing, 3D rapid prototyping, etc.). Examples of additive manufacturing techniques include fused deposition modeling, electron beam melting, laminated object manufacturing, selective laser sintering (including direct metal laser sintering, also known as direct metal laser melting or selective laser melting), digital light processing, and stereolithography, among others. Despite the continued advancement of 3D printing technology, the process of building structures layer by layer is complex, inefficient, and prone to failure. Errors in the 3D process can result in weaknesses or failures in manufactured parts, resulting in waste, risk, and other unreliability.

[0020] The phrase "additive manufacturing device" is used interchangeably herein with the phrase "printing device" and the term "printer", and the term "printing" is used interchangeably herein with the word "building", referring to the action of building a structure using an additive manufacturing device, regardless of the specific additive manufacturing technology used to form the structure. As used herein, printing refers to various forms of additive manufacturing, and includes three-dimensional (3D) printing or 3D rapid prototyping, as well as sintering or melting / melting techniques. For example, an additive manufacturing system can use an electron beam or a laser beam to manufacture a build. An additive manufacturing system may include multiple electron beam guns or laser designs. Examples of additive manufacturing or printing techniques include, among others, fused deposition modeling, electron beam melting, laminated object manufacturing, selective laser sintering (including direct metal laser sintering, also known as direct metal laser melting or selective laser melting), and stereolithography.

[0021] For example, selective laser melting (SLM), also known as direct metal laser melting (DMLM), direct metal laser sintering (DMLS), or laser powder bed fusion (LPBF), is a rapid prototyping, 3D printing, or additive manufacturing (AM) technology designed to melt and fuse metal powders together using a high power density laser. For example, the SLM process can completely melt metal materials into solid three-dimensional parts.

[0022] SLM is a part of additive manufacturing in which high-power density lasers are used to melt and fuse metal powders together. With SLM, a thin layer of atomized fine metal powder is evenly distributed on a substrate (e.g., metal, etc.) using a coating mechanism. The substrate is fastened to an indexing table that moves in a vertical (Z) axis. This occurs in a chamber containing a strictly controlled atmosphere of an inert gas (e.g., argon or nitrogen with an oxygen level below 500 parts per million). Once each layer has been distributed, each two-dimensional (2D) slice of the part geometry is melted by selectively melting the powder. The melting of the powder is accomplished by a high-power laser beam (such as a hundreds of watts of ytterbium (Yb) fiber laser, etc.). The laser beam is directed in the X and Y directions by two high-frequency scanning mirrors. The laser energy is strong enough to allow complete melting (welding) of the particles to form solid metal. The process is repeated layer by layer until the part is complete.

[0023] Direct Metal Laser Melting (DMLM) or Direct Metal Laser Sintering (DMLS) are special types of SLM that use a variety of alloys and allow prototypes to become functional hardware made from the same material as production parts. Because parts are built layer by layer, it is possible to design organic geometries, internal features and challenging passages that cannot be cast or otherwise machined. For example, DMLS produces strong and durable metal parts that can, for example, work well as functional prototypes and / or end-use production parts.

[0024] The object is built directly from the file generated from the CAD (computer-aided design) data. The DMLS process begins by slicing the 3D CAD file data into layers (e.g., from 20 to 100 microns thick, 30-120 μm thick, 50-150 μm thick, etc.), creating a two-dimensional (2D) image of each layer. For example, the format of the 3D CAD file is a .stl file used for most layer-based 3D printing or stereolithography technologies. The file is then loaded into a file preparation software package, which assigns parameters, values, and physical supports that allow the file to be interpreted and built by, for example, different types of additive manufacturing machines.

[0025] In DMLS / DMLM, a laser is used to selectively melt thin layers of fine particles to produce objects that exhibit fine, dense and uniform properties. For example, a DMLS machine uses a high-power 200-watt Yb fiber laser. The machine includes a build chamber area, which includes a material dispensing platform and a build platform, as well as a recoater blade for moving new powder on the build platform. The technology melts metal powder into a solid part by locally melting it using a focused laser beam. When the powder melts due to exposure to laser beam radiation, a molten pool is formed. The part is additively built layer by layer (for example, using layers that are 10 microns thick, 20 μm thick, 30 μm thick, 50 μm thick, etc.).

[0026] The DMLS process begins with a roller spreading a thin layer of metal powder over the print bed. Next, the laser is guided based on the CAD data to create a cross-section of the object by completely melting the metal particles. The print bed is then lowered so that the process can be repeated to create the next layer of the object. After all layers are printed, the excess unmelted powder is brushed, scraped or blown off. The object typically requires little to no finishing.

[0027] For example, the machine may include and / or operate with monitoring and control systems and methods (such as iterative learning control, continuous auto-calibration, and real-time melt pool monitoring, etc.) to introduce step changes in build process performance and stability. Certain examples implement melt pool monitoring, iterative learning control, continuous auto-calibration, real-time melt pool control, filter monitoring, pump monitoring, spray application monitoring, cathode and beam monitoring, etc.

[0028] Other additive manufacturing methods such as electron beam melting (EBM) can be used for metal alloys that are prone to cracking, such as titanium. With EBM, a high-performance electron beam source and in-situ process monitoring using "self-generated" x-ray imaging and backscattered electron technology can be used to improve quality control. Binder jetting allows for rapid printing at a lower cost, with novel support structure designs and clean burning binders to address the two key technical challenges of sintering deformation and material properties, thereby enabling additive manufacturing for automobiles, other transportation solutions, powder generation, etc. EBM utilizes raw materials in the form of metal powders or metal wires, which are placed under vacuum (e.g., in a vacuum-sealed build chamber). Generally speaking, the raw materials are melted together by heating via an electron beam.

[0029] Systems utilizing EBM typically acquire data from a 3D computer-aided design (CAD) model and use the data to place successive layers of raw material using a device that spreads the raw material, such as a powder distributor. Successive layers are melted together using a computer-controlled electron beam. As described above, the process is performed under vacuum within a vacuum-sealed build chamber, which makes the process suitable for manufacturing parts using reactive materials that have a high affinity for oxygen (e.g., titanium). In some examples, the process operates at higher temperatures (up to about 1200°C) relative to other additive manufacturing processes, which can result in differences in phase formation through solidification and solid-state phase transformations.

[0030] The examples described herein are applicable to other additive manufacturing methods that use other types of additive manufacturing devices other than the additive manufacturing devices disclosed herein. For example, directed energy deposition (DED), direct ceramics, BinderJet, stereolithography, photopolymerization, etc. can benefit from the monitoring, analysis, and correction described herein.

[0031] "Include" and "comprising" (and all their forms and tenses) are used herein as open-ended terms. Thus, whenever a claim adopts any form of "include" or "comprising" (e.g., includes, comprises, has, etc.) as a preamble or in any type of claim statement, it should be understood that additional elements, terms, etc. may be present without exceeding the scope of the corresponding claim or statement. As used herein, when the phrase "at least" is used as a transitional term, such as in the preamble of a claim, it is open-ended in the same manner as the terms "include" and "comprising" are open-ended. The term "and / or" when used, for example, in a form such as A, B, and / or C, refers to any combination or subset of A, B, C, such as (1) A alone, (2) B alone, (3) C alone, (4) A and B, (5) A and C, (6) B and C, and (7) A and B and C. As used herein in the context of describing structures, components, items, objects, and / or things, the phrase "at least one of A and B" is intended to refer to any of (1) at least one A, (2) at least one B, and (3) at least one A and at least one B. Similarly, as used herein in the context of describing structures, components, items, objects, and / or things, the phrase "at least one of A or B" is intended to refer to any of (1) at least one A, (2) at least one B, and (3) at least one A and at least one B. As used herein in the context of describing the performance or execution of processes, instructions, actions, activities, and / or steps, the phrase "at least one of A and B" is intended to refer to any of (1) at least one A, (2) at least one B, and (3) at least one A and at least one B. Similarly, as used herein in the context of describing the performance or execution of processes, instructions, actions, activities, and / or steps, the phrase "at least one of A or B" is intended to refer to any of (1) at least one A, (2) at least one B, and (3) at least one A and at least one B.

[0032] As used herein, singular references (e.g., "a", "a", "first", "second", etc.) do not exclude the plural. As used herein, the term "a" or "an" entity refers to one or more of that entity. The terms "a" (or "a"), "one or more", and "at least one" are used interchangeably herein. In addition, although listed separately, multiple devices, elements, or method actions may be implemented by, for example, a single unit or processor. In addition, although individual features may be included in different examples or claims, these may potentially be combined, and the inclusion in different examples or claims does not mean that the combination of features is not feasible and / or disadvantageous.

[0033] As used herein, the terms "system," "unit," "module," "engine," "component," and the like may include hardware and / or software systems that operate to perform one or more functions. For example, a module, unit, or system may include a computer processor, a controller, and / or other logic-based devices that operate based on instructions stored on a tangible and non-transitory computer-readable storage medium (e.g., a computer memory). Alternatively, a module, unit, or system may include a hard-wired device that operates based on the hard-wired logic of the device. The various modules, units, engines, and / or systems shown in the figures may represent hardware that operates based on software or hard-wired instructions, software that directs hardware to operate, or a combination thereof.

[0034] The present disclosure generally relates to devices, systems and methods for monitoring, determining and adjusting the health of an additive manufacturing device (also referred to as an additive manufacturing machine), a group of additive manufacturing devices and / or associated processes, builds, etc. For example, a build involves forming a part (e.g., a blade, a rotor, a stator, a trunnion, a housing, a shroud, other industrial parts, etc.) by an additive manufacturing machine. Example systems and methods acquire data of a build, process, machine configuration, etc., and ingest, fuse, analyze and aggregate the data for state determination and analysis. For example, these methods may include considering and analyzing actual sensor data from a machine or process, as well as metadata derived from statistical process control and / or other quality scores. Certain examples determine the health of a build, process and / or machine on a hierarchical and / or aggregate basis without the need for manual analysis. In addition, the systems and methods according to the present disclosure improve the accuracy of diagnosing a build and / or additive manufacturing device to provide an actionable output (e.g., to correct the build, adjust the additive manufacturing device and / or associated processes, etc.).

[0035] Certain examples provide apparatus, systems, and methods for analyzing and evaluating behavioral changes within a group of additive manufacturing machines. An automated system is capable of learning behavioral changes of the group and identifying non-standard builds. Certain examples may be deployed on an additive manufacturing machine and / or separate from an additive manufacturing machine. When deployed on an additive manufacturing machine, the system / method enables detection of critical / non-standard builds by the additive manufacturing machine while a part is being printed and / or immediately after a part is printed to anticipate scrap, foresee ad-hoc incremental control / inspection, and enable collective action of the entire group to contain the changes and improve production yield.

[0036] In some examples, an analyzer system (e.g., a standard behavior analyzer or SBA) is configured to read parameters and / or features representing various processes performed by an additive manufacturing machine during the construction of a part. The analyzer system includes a learner subsystem circuit and an evaluator subsystem circuit. For example, the learner subsystem reads log files and / or other data (e.g., control, sensor and / or subsystem data from machine components, etc.) of a given build group, and learns the behavior of the associated additive manufacturing machine by performing a relative or comparative analysis of the build group using unsupervised machine learning and statistical methods. This analysis can identify non-standard builds within the analyzed build group and determine the machine parameters that cause non-standard behavior (and standard behavior). As used herein, standard behavior indicates a build and / or related process, additive manufacturing machine configuration, etc. that meets the expected results (e.g., target or minimum part quality, etc.), acceptable build output, other expected results, etc. Non-standard behavior indicates a build and / or related process, additive manufacturing machine configuration, etc. that does not meet (e.g., does not meet or exceed, etc.) the expected results, acceptable build output, other expected results, etc.

[0037] However, in certain examples disclosed herein, non-standard may also indicate anomalous behavior that is not necessarily "bad" or of insufficient quality. Instead, non-standard processing behavior may also be due to leaner / optimized / better processing introduced by new / fine-tuned parameters or settings or methods. Thus, an anomalous build does not necessarily indicate a suboptimal build. The Additive Manufacturing development process relies on and benefits from in-situ quality assurance techniques, not only for corrective purposes, but also for identifying opportunities for (continuous) improvement. Thus, a non-standard build identified based on anomaly detection does not itself necessarily indicate a "defect" in the build.

[0038] In some examples, the behaviors (standard and / or non-standard) learned by the learner subsystem are saved in digital form for later use by the evaluator subsystem. New builds can then be evaluated based on the learned behaviors of the saved build population to assess whether the new build is non-standard, quantify the behavior of the new build, and identify one or more machine parameters, build settings, etc. that cause such (non-standard or standard) behavior. Differences from standard build / machine behavior can include critical / important data such as build preparation, setup / configuration, and facility inputs (e.g., facility gas / water / electricity), and can be evaluated at the machine level, subsystem level (e.g., beam subsystem, etc.), etc. The associated processing can be automated and adopted with respect to various build populations (e.g., all builds in the past six months, all builds of a given type in the past three months, all builds on a given additive manufacturing machine type in the past year, etc.). Such systems and methods can be implemented for any additive manufacturing machine population (DMLM, EBM, BinderJet, etc.).

[0039] Using the systems, devices, and methods described herein, build populations may be used to identify non-standard and / or otherwise abnormal additive manufacturing machine behavior, and such behavior may be quantified / measured with metrics to enable comparison with other build / machine behaviors, identification of parameter-level contributions, and thereby determination of one or more of the most significant / most influential parameters that cause the non-standard behavior. For example, identification of parameters, settings, etc. enables specific actions to be taken to contain changes in machine behavior, thereby reducing the number / amount of quality defects caused by such changes. Such analysis may also reduce or eliminate "bad WIP" (work in progress) from entering expensive post-production processing (e.g., stress relief, heat treatment, computed tomography (CT) inspection, etc.), and / or may inform such processing such as inspection recommendations (changes identified in layers 1xx-2yy), etc.

[0040] Parameter limits defining standard behavior can be learned from a given build population. Standard behavior can be learned at the build level and at the layer level (e.g., by evaluating the layers in each build population (e.g., standard melting times for the layers, etc.)). For example, a layered evaluation of standard behavior enables identification of specific layers that contribute to non-standard behavior / outcomes. Anomalous layers can be identified in a build when compared to a build population (e.g., layers with anomalous melting times, etc.), thereby being part of the build-level non-standard behavior.

[0041] In some examples, the use of unsupervised learning does not necessarily imply that standard vs. non-standard classifications match good / critical quality. However, the unsupervised learning model provides a mechanism for isolating non-standard processing behavior that supports pre-emptive identification and possible containment of anomalous, abnormal, or otherwise incorrect processing behavior, such as through recommended corrective actions to stop / adjust processing variability that, if not corrected, may result in systematic or random undesirable processing outcomes. In some examples, classifying a build as standard or non-standard, and associating the classification with an indication of good or critical quality includes comparing combined quality data (e.g., CT, etc.) with a complementary supervised learning step to fine-tune the classification algorithm to, for example, further differentiate within a standard build group or layer or feature range to identify what is "good" from a quality perspective.

[0042] Thus, critical builds and / or associated AM machines can be identified immediately after a part has been printed. This identification enables anticipation of scrap, prediction of interim incremental control / inspection, etc. This identification can also be done layer by layer, leveraging information from the analytical processor, etc., when comparing layers to a reference / golden build.

[0043] During each population build, additive manufacturing machine behavior can be represented by a set of powerful domain-based features, SPC quality scores, etc. A given build population is analyzed using a novel combination of unsupervised machine learning methods (e.g., k-NN anomaly detection + HDBScan clustering, etc.), and statistical univariate models are employed to evaluate and find critical / non-standard builds, as well as operational limits for standard build / machine behavior. These methods are combined to capture nonlinear interactions of features and their impact on build / machine behavior. For example, changes can be measured at the feature level. In addition, a novel aggregate metric (referred to herein as non-compliance severity) is designed to measure changes at the build and feature levels. The non-compliance severity metric identifies key features that contribute to non-standard behavior. For example, learned population behaviors can be saved in a database and deployed to enable detection on new builds.

[0044] Certain examples provide methods for classifying machine behavior during a build in an additive manufacturing machine by integrating and / or decomposing build data (e.g., power usage, ambient temperature, etc.), layer data (e.g., melt pool melt temperature, vacuum level, etc.), and machine data (e.g., settings of the additive manufacturing machine, etc.) into aggregate build features to introduce a high-fidelity machine processing-based representation of machine behavior during the build. Nonlinear (e.g., clustering (DBScan), neural networks, random forests, k-means, etc.), linear (Mahalanobis distance, Hotelling T, control charts, etc.), multivariate and / or univariate methods can be used alone or in various combinations to classify the build as standard or critical / nonstandard and further perform build-level behavior / quality estimation and feature-level behavior estimation. Non-Compliance Severity (NCS) is an aggregate metric used to measure / quantify changes in the behavior of an additive manufacturing machine during a build. The NCS metric can be further decomposed to the feature level to establish possible causal relationships between features, which enables the identification of features that are most helpful in identifying the behavior. The NCS metric can be used for compliance and problem solving of parts built on additive manufacturing machines.

[0045] Certain examples provide a system for classifying and reporting machine behavior in an additive manufacturing machine. The system includes an automatic learner subsystem deployed in a shared computer connected to multiple additive manufacturing machines. The automatic learner subsystem can read a build population represented by features. The automatic learner subsystem learns behavior from a build population of one or more additive manufacturing machines, and can classify each build and measure the behavior of each build. The automatic learner subsystem can learn hierarchical behavior variations, can learn online and offline modes at a desired frequency, and can learn from real and / or simulated data of additive manufacturing machines. The automatic learner subsystem can report the behavior classification of each build, features that contribute to the behavior, limitations of the features, hierarchical behavior, etc., and can generate interactive displays and generate actions that can be used to reduce behavior variations. For example, the learned behavior can be saved as a reference behavior in a data storage device (e.g., for future evaluation of new builds).

[0046] The system also includes an automatic evaluator subsystem. For example, the automatic evaluator subsystem can be deployed in each additive manufacturing machine. The automatic evaluator system can read a new build represented by features and evaluate the new build based on one or more reference / learned behaviors to classify the behavior associated with the new build. For example, the build can be evaluated at the build level and the layer level. The evaluation results can be generated and reported immediately after the build is completed. For example, the report can detail the classification of layers and / or builds as standard / non-standard and the features that contribute to the classification, and the report can trigger temporary incremental controls, inspections, other adjustments to correct for non-standard behaviors. In some examples, sensitivity can be set (e.g., automatically and / or manually via sliders, ranges, values, etc.) to set certain thresholds to higher, lower, specific custom ranges / tolerances, etc.

[0047] Figure 1 An additive manufacturing apparatus of the present disclosure is depicted according to one or more of the examples shown and described herein. Figure 1 As shown, additive manufacturing system 100 (also referred to herein as additive manufacturing apparatus, additive manufacturing machine, additive machine, etc.) includes at least a build chamber 102, an imaging device 114, and a control component 120. Build chamber 102 defines an interior 104, which is separated from an external environment 105 via one or more chamber walls 103. In some examples, at least a portion of one or more chamber walls 103 of build chamber 102 may include a window 106 therein. Imaging device 114 is generally located in external environment 105, adjacent to build chamber 102 (i.e., not within interior 104 of build chamber 102), and is arranged such that a field of view 116 of imaging device 114 extends through window 106 into interior 104 of the chamber.

[0048] In some examples, interior 104 of build chamber 102 can be a vacuum-sealed interior such that article 142 formed within build chamber 102 is formed under optimal conditions for EBM or DMLM, as is generally understood. Build chamber 102 can maintain a vacuum environment via a vacuum system. As is generally understood, the vacuum system shown can include, but is not limited to, a turbomolecular pump, a vortex pump, an ion pump, and one or more valves. In some examples, the vacuum system can be communicatively coupled to control component 120 such that control component 120 directs the operation of the vacuum system to maintain a vacuum within interior 104 of build chamber 102. In some examples, the vacuum system can maintain a vacuum of approximately 1×10 0.1% throughout the build cycle. -5 mbar or lower base pressure. In a further example, the vacuum system may provide approximately 2×10 -3 Partial pressure of helium or other reactive or inert control gas in mbar.

[0049] In other examples, build chamber 102 can be disposed in a closable chamber provided with ambient air and atmospheric pressure. In other examples, build chamber 102 can be disposed in the open air.

[0050] Build chamber 102 generally includes within interior 104 a powder bed 110 supporting a powder layer 112 thereon and a powder distributor 108. In some examples, build chamber 102 may also include one or more raw material hoppers 140a, 140b in which raw material 141 is maintained. In some examples, build chamber 102 may also include a launcher 130. Build chamber 102 may also include other components, particularly components that facilitate EBM or DMLM, including components not specifically described herein.

[0051] The powder bed 110 is generally a platform or container located within the interior 104 of the build chamber 102, which is arranged to receive raw material 141 from one or more raw material hoppers 140a, 140b. The size or configuration of the powder bed 110 is not limited by the present disclosure, but can generally be shaped and sized to accommodate a certain amount of raw material 141 from the raw material hoppers 140a, 140b in the form of a powder layer 112, one or more portions of an article 142, and / or unmelted raw material 141, as described in more detail herein.

[0052] In some examples, powder bed 110 may include a movable building platform 111 supported by a lifting component 113. Movable building platform 111 may generally be a surface within powder bed 110 that can be moved in a vertical direction (e.g., in a vertical direction) of the system by lifting component 113. Figure 1The movable building platform 111 in the powder bed 110 can be moved in the +y / -y direction of the coordinate axis of the powder bed 110 to increase and / or decrease the total volume of the powder bed 110. For example, the movable building platform 111 in the powder bed 110 can be moved in the downward direction (e.g., toward the Figure 1 The movable build platform 111 may be moved in the -y direction of the coordinate axis of the matrix 140 to increase the volume of the powder bed 110. In addition, the movable build platform 111 may be moved by a lifting member 113 to add each successive powder layer 112 to the object 142 being formed, as described in more detail herein.

[0053] Lifting member 113 is not limited by the present disclosure and can generally be coupled to movable building platform 111 and movable to move the system in a vertical direction (e.g., in a vertical direction). Figure 1 102 ). In some examples, lifting component 113 may utilize a linear actuator type mechanism to achieve movement of movable build platform 111. Illustrative examples of devices or systems suitable for use as lifting component 113 include, but are not limited to, a scissor lift, a mechanical linear actuator (such as a screw-based actuator), an axle actuator (e.g., a rack and pinion actuator), a hydraulic actuator, a pneumatic actuator, a piezoelectric actuator, an electromechanical actuator, and the like. In some examples, lifting component 113 may be located within build chamber 102. In other examples, lifting component 113 may be located only partially within build chamber 102, particularly in examples where it may be desirable to isolate portions of lifting component 113 that are sensitive to harsh conditions (high heat, excessive dust, etc.) within interior 104 of build chamber 102.

[0054] Powder distributor 108 is generally arranged and configured to lay and / or spread a layer of raw material 141 as powder layer 112 in powder bed 110 (e.g., on a starting plate or building platform 111 within the powder bed). That is, powder distributor 108 is arranged so that movement of powder distributor 108 is controlled by Figure 1 For example, the powder distributor 108 may be located on or above the powder bed 110. Figure 1 108 may include an arm, rod, or the like extending a certain distance in the z direction of the coordinate axis of the powder bed 110 (e.g., from the first end to the second end of the powder bed 110). In some examples, the length of powder distributor 108 may be longer than the width of build platform 111 so that powder layer 112 may be distributed at each position of build platform 111. In some examples, powder distributor 108 may have a top surface that is parallel to (e.g., approximately parallel to) the top surface of build platform 111. Figure 1One or more motors, actuators, etc. may be coupled to the powder distributor 108 to achieve movement of the powder distributor 108. For example, a rack and pinion actuator may be coupled to the powder distributor 108 to move the powder distributor 108 in a certain direction. Figure 1 Move back and forth on the powder bed in the +x / -x direction of the coordinate axis, such as Figure 1 . In some examples, the movement of the powder distributor 108 can be continuous (e.g., moving without stopping except for changing direction). In other examples, the movement of the powder distributor 108 can be stepwise (e.g., moving in a series of intervals). In other examples, the movement of the powder distributor 108 can be such that multiple interruptions occur between periods of movement.

[0055] As described in more detail herein, the powder distributor may further include one or more teeth (e.g., rake fingers, etc.) extending from the powder distributor 108 into the raw material 141 in the raw material hoppers 140a, 140b, thereby disrupting the raw material 141 (e.g., distributing the raw material 141, spreading the powder layer 112, etc.) as the powder distributor 108 moves.

[0056] In some examples, the powder distributor 108 includes a bottom surface B extending from the powder distributor 108 (e.g., generally toward Figure 1 In some examples, rake teeth 107 may be arranged substantially perpendicular to a plane of building platform 111 (e.g., perpendicular to a plane formed by a plurality of rake teeth 107 extending in the -y direction of the coordinate axis of the y coordinate system). Figure 1 In another example, rake teeth 107 can be tilted relative to building platform 111. The angle a of tilted rake teeth 107 relative to the normal of the building platform can be any value, and in some examples, angle a is between about 0° and about 45°.

[0057] In some examples, each of plurality of tines 107 may be a metal foil or sheet. The total length of plurality of tines 107 may be longer than the width of build platform 111 so that powder can be distributed at every location of build platform 111. The shape and size of tines 107 may be designed to rake through raw material 141 to distribute powder layer 112 on build platform 111. Some examples may not include tines 107.

[0058] It should be understood that although the powder distributor 108 described herein is generally Figure 1 The coordinate axis depicted in the figure extends a certain distance in the x direction and Figure 1112 to spread the powder layer 112 as described above, but this is only an illustrative example. Other configurations are also contemplated. For example, the powder distributor 108 may rotate about an axis to spread the powder layer 112, may be articulated about one or more joints, etc. to spread the powder layer 112, etc., without departing from the scope of the present disclosure.

[0059] In some examples, the cross-section of the powder distributor 108 may be generally triangular, such as Figure 1 However, it should be understood that the cross-section can be of any shape, including but not limited to circular, elliptical, square, rectangular, polygonal, etc. The height of the powder distributor 108 can be set so that the powder distributor 108 is in the vertical direction of the system (e.g., along the Figure 1 The height of the powder distributor 108 can also be selected to provide specific mechanical strength to the powder distributor 108 (the +y / -y axis of the coordinate axis of the system). That is, in some examples, the powder distributor 108 can have a specific controllable deflection in the vertical direction of the system. Taking into account that the powder distributor 108 pushes a certain amount of raw material 141, the height of the powder distributor 108 can also be selected. If the height of the powder distributor 108 is too small, the powder distributor 108 can only push forward a smaller amount relative to a higher power powder distributor 108. However, if the height of the powder distributor 108 is too high, the powder distributor 108 may complicate the capture of powder from the powder sieve (for example, the higher the height of the powder distributor 108, the greater the force may be required to capture a predetermined amount of powder from the powder sieve by moving the powder distributor 108 into the powder sieve and allowing a predetermined amount of powder to fall on the top of the powder distributor 108 from a first side in the direction of travel into the powder sieve to a second side in the direction of the build platform 111). In still other examples, the height of the powder distributor 108 may be such that areas adjacent to the leading edge and the trailing edge of the powder distributor 108 are within the field of view 116 of the imaging device 114, as described herein.

[0060] In some examples, the powder distributor 108 can be communicatively coupled to the control component 120, such as Figure 1 As used herein, the term "communicatively coupled" generally refers to any link in a manner that facilitates communication. Therefore, "communicatively coupled" includes wireless and wired communications, including wireless and wired communications now known or later developed. When the powder distributor 108 is communicatively coupled to the control component 120, the control component 120 can transmit one or more signals, data, etc. to cause the powder distributor 108 to move, change direction, change speed, etc. For example, a "reverse" signal transmitted by the control component 120 to the powder distributor 108 can cause the powder distributor 108 to reverse its direction of movement (e.g., from movement in the +x direction to movement in the -x direction).

[0061] Each of the raw material hoppers 140a, 140b may generally be a container that holds a certain amount of raw material 141 therein and includes an opening to dispense the raw material 141 therefrom. Figure 1 Two raw material hoppers 140a, 140b are depicted, but the present disclosure is not limited thereto. That is, any number of raw material hoppers may be utilized without departing from the scope of the present disclosure. Figure 1 Raw material hoppers 140a, 140b are depicted as being located within interior 104 of build chamber 102, but the present disclosure is not limited thereto. That is, in various other examples, raw material hoppers 140a, 140b may be located or partially located outside of build chamber 102. However, it should be understood that if the raw material hoppers are located or partially located outside of build chamber 102, one or more outlets of the raw material hoppers supplying raw material 141 may be selectively sealed when raw material 141 is not being distributed to maintain a vacuum within build chamber 102.

[0062] The shape and size of the raw material hoppers 140a, 140b are not limited by the present disclosure. That is, without departing from the scope of the present disclosure, the raw material hoppers 140a, 140b can generally have any shape and / or size. In some examples, the shape and / or size of each of the raw material hoppers 140a, 140b can be designed to conform to the size of the build chamber 102, so that the raw material hoppers 140a, 140b can be adapted to fit within the build chamber. In some examples, the shape and size of the raw material hoppers 140a, 140b can be designed so that the total volume of the raw material hoppers 140a, 140b is sufficient to accommodate a certain amount of raw material 141 required to make an article 142, and the certain amount of raw material 141 includes a sufficient amount of material to form each continuous powder layer 112 and additional material constituting the unmelted raw material 141.

[0063] The raw material hoppers 140a, 140b may generally have an outlet for ejecting the raw material 141 within the raw material hoppers 140a, 140b so that the raw material 141 may be dispersed by the powder distributor 108, as described herein. Figure 1), the raw material 141 can flow freely out of the raw material hoppers 140a, 140b under the action of gravity, thereby forming a layer or pile of raw material 141 for distribution by the powder distributor 108. In other examples, the outlets of the raw material hoppers 140a, 140b can be selectively closed via a selective closing mechanism so that only a portion of the raw material 141 located in the respective raw material hoppers 140a, 140b is distributed at a particular time. The selective closing mechanism can be communicatively coupled to the control component 120 so that data and / or signals transmitted to / from the control component 120 can be used to selectively open and close the outlets of the raw material hoppers 140a, 140b.

[0064] The raw material 141 contained in the raw material hoppers 140a, 140b and used to form the article 142 is not limited by the present disclosure and can generally be any raw material for EBM or DMLM now known or later developed. Illustrative examples of the raw material 141 include, but are not limited to, pure metals such as titanium, aluminum, tungsten, etc.; and metal alloys such as titanium alloys, aluminum alloys, stainless steel, cobalt-chromium alloys, cobalt-chromium-tungsten alloys, nickel alloys, etc. Specific examples of the raw material 141 include, but are not limited to, Ti6Al4V titanium alloy, Ti6Al4V ELI titanium alloy, Grade 2 titanium, and ASTM F75 cobalt-chromium alloy (all available from Sweden) Another specific example of raw material 141 is available from Specialty Metals, Inc. (Huntington WV). Alloy 718.

[0065] In some examples, the raw material 141 is pre-alloyed, rather than a mixture. This may allow EBM or DMLM to be classified with selective laser melting (SLM), where other techniques such as selective laser sintering (SLS) and direct metal laser sintering (DMLS) require heat treatment after fabrication. Compared to selective laser melting (SLM) and DMLS, EBM has a generally higher build rate due to its higher energy density and scanning method.

[0066] The emitter 130 is typically a device that emits an electron beam (e.g., a charged particle beam), such as an electron gun, a linear accelerator, etc. The emitter 130 generates an energy beam 131, which can be used to melt or fuse the raw material 141 together when the raw material 141 is spread as a powder layer 112 on the building platform 111. In some examples, the emitter 130 may include at least one focusing coil, at least one deflection coil, and an electron beam power supply, which may be electrically connected to the emitter control unit. In an illustrative example, the emitter 130 generates a focusable electron beam having an acceleration voltage of about 60 kilovolts (kV) and a beam power in the range of about 0 kilowatts (kW) to about 10 kW. When the article 142 is built by melting each successive powder layer 112 using the energy beam 131, the pressure in the vacuum chamber may be about 1×10 -3 mBar is about 1×10 -6 The emitter 130 may be located in a gun vacuum chamber. The pressure in the gun vacuum chamber may be about 1×10 -4 mBar to about 1×10 -7 mBar range. In some examples, the emitter 130 can emit a laser beam using direct metal laser melting (DMLM). The emitter 130 can emit a laser to melt ultra-thin layers of metal powder to build a three-dimensional object. When using DMLM, a gas flow can be provided over the build compared to electron beam melting manufacturing that requires a vacuum chamber.

[0067] In some examples, the transmitter 130 can be communicatively coupled to the control component 120, such as Figure 1 1 and 120. The communication connection between the transmitter 130 and the control component 120 may provide the ability to transmit signals and / or data (such as control signals from the control component 120 that direct the operation of the transmitter 130) between the transmitter 130 and the control component 120.

[0068] Still reference Figure 1, the imaging device 114 is typically located in the external environment 105 outside the build chamber 102, but is positioned so that the field of view 116 of the imaging device 114 passes through the window 106 of the build chamber 102. The imaging device 114 is typically positioned outside the build chamber 102 so that the harsh environment within the interior 104 of the build chamber 102 does not affect the operation of the imaging device 114. That is, heat, dust, metallization, x-ray radiation, etc. occurring within the interior 104 of the build chamber 102 do not affect the operation of the imaging device 114. In some examples, the imaging device 114 is fixed in an appropriate position so that the field of view 116 remains constant (e.g., does not change). In addition, the imaging device 114 is arranged in a fixed position so that the field of view 116 of the imaging device 114 covers the entire powder bed 110. That is, the imaging device 114 is able to image the entire powder bed 110 within the build chamber 102 through the window 106.

[0069] In some examples, imaging device 114 is a device that is specifically configured to sense electromagnetic radiation, particularly thermal radiation (e.g., heat energy radiation) generated by various components within powder bed 110 (e.g., powder layer 112, raw material 141, and / or article 142). Thus, imaging device 114 may generally be a device that is specifically tuned or otherwise configured to obtain images in a spectrum where thermal radiation is readily detected, such as the visible spectrum and the infrared spectrum (including the far infrared and near infrared spectrum). Thus, an illustrative example of a device that is specifically tuned or otherwise configured to obtain images in a spectrum where thermal radiation is readily detected includes, but is not limited to, an infrared camera. In some examples, imaging device 114 may be a camera sensitive in a wavelength range of about 1 micrometer (μm) to about 14 μm, including about 1 μm, about 2 μm, about 3 μm, about 4 μm, about 5 μm, about 6 μm, about 7 μm, about 8 μm, about 9 μm, about 10 μm, about 11 μm, about 12 μm, about 13 μm, about 14 μm, or any value or range between any two of these values ​​(including endpoints). Thus, imaging device 114 is suitable for imaging the temperatures occurring during EBM or DMLM of powder layer 112. In some examples, the wavelength sensitivity of imaging device 114 may be selected depending on the type of raw material used. Illustrative examples of suitable devices that may be used for imaging device 114 include, but are not limited to, IR cameras (infrared cameras), NIR cameras (near infrared cameras), VISNIR cameras (visual near infrared cameras), CCD cameras (charge coupled device cameras), line scan cameras, and CMOS cameras (complementary metal oxide semiconductor cameras).

[0070] In some examples, the imaging device 114 can be an area scan camera that is capable of providing data specific to one or more regions of interest within the field of view 116 (including regions of interest that move within the field of view 116). That is, the area scan camera includes a pixel matrix that allows the device to capture a 2D image with vertical and horizontal elements within a single exposure cycle. The area scan camera can also be used to obtain multiple consecutive images, which is useful when selecting a region of interest within the field of view 116 and observing changes in the region of interest, as described in more detail herein. Illustrative examples of such area scan cameras include those available from Basler AG (Ahrensburg, Germany), JAI Co., Ltd. (Yokohama, Japan), National Instruments (Austin, Texas), and Stemmer Imaging (Puchheim, Germany). In some examples, the imaging device 114 can be a line scan camera, which can be used for jet pattern detection and / or power deposition anomaly detection of short feeds, agglomerations, pushes, etc.

[0071] In some examples, imaging device 114 may have a monochrome image sensor. In other examples, imaging device 114 may have a color image sensor. In various examples, imaging device 114 may include one or more optical elements, such as lenses, filters, etc. In a specific example, imaging device 114 may include a Bayer filter. As is well known, a Bayer filter is a color filter array (CFA) that arranges RGB filters on a square grid of a light sensor to create a color image, such as a filter pattern of approximately 50% green, approximately 25% red, and approximately 25% blue.

[0072] In some examples, imaging device 114 may also be a device specifically configured to provide signals and / or data corresponding to the sensed electromagnetic radiation to control component 120. Thus, imaging device 114 may be communicatively coupled to control component 120, such as Figure 1 As shown by the dotted line drawn between the imaging device 114 and the control component 120.

[0073] It should be appreciated that by positioning imaging device 114 in external environment 105 outside interior 104 of build chamber 102, an existing build chamber having a window in chamber wall 103 may be readily retrofitted with a kit including imaging device 114 to upgrade the existing build chamber with the capabilities described herein.

[0074] The control component 120 (also referred to as an additive machine controller) is generally a device that is communicatively coupled to one or more components of the additive manufacturing system 100 (e.g., the powder distributor 108, the imaging device 114, and / or the emitter 130), and is particularly arranged and constructed to transmit signals and / or data to and / or receive signals and / or data from one or more components of the additive manufacturing system 100 (such as the imaging device 114, one or more sensors 150-151 (residual oxygen percentage sensor, laser track temperature sensor, dew point temperature sensor, heating temperature sensor, differential pressure sensor, etc.) positioned relative to the additive manufacturing system 100 (e.g., positioned on or within a component of the additive manufacturing system 100) to record temperature, motion, vibration, power, etc.

[0075] Figure 2 2 is a block diagram of an example architecture 200 according to one or more examples shown and described herein. In some examples, the architecture or infrastructure 200 may include an additive manufacturing machine 100, a server 210, a user computing device 220, and a mobile computing device 230. For example, one or more of the server 210, the user computing device 220, or the mobile computing device 230 may implement an additive machine controller 120. The additive manufacturing machine 100 (also referred to herein as an additive machine, an additive manufacturing system, an additive manufacturing device, and / or an additive manufacturing apparatus) may be communicatively coupled to the server 210, the user computing device 220, and the mobile computing device 230 via a network 240. In some examples, the network 240 may include one or more computer networks (e.g., personal area networks, local area networks, or wide area networks), cellular networks, satellite networks, and / or global positioning systems, and combinations thereof. Thus, the user computing device 220 may be communicatively coupled to the network 240 via a wide area network, via a local area network, via a personal area network, via a cellular network, via a satellite network, and the like. Suitable local area networks may include wired Ethernet and / or wireless technologies such as Wireless Fidelity (Wi-Fi). Suitable personal area networks may include wireless technologies such as IrDA, Wireless USB, Z-Wave, ZigBee and / or other near field communication protocols. Suitable cellular networks include, but are not limited to, technologies such as LTE, WiMAX, UMTS, CDMA, and GSM.

[0076] In some examples, the additive manufacturing system 100 may transmit captured information related to the build, such as images, sensor signals (e.g., Open Platform Connectivity Unified Architecture (OPC UA) signals, etc.), build status, log files, etc., to the server 210, the user computing device 220, and / or the mobile computing device 230. The log files may include multiple parameters output from multiple subsystems of the additive manufacturing system 100 (such as a vacuum system, a beam system, a powder layering system, etc.). The multiple parameters may be raw data output from the additive manufacturing system 100, or parameters that are further processed based on machine operation. For example, the parameters may be processed based on domain knowledge and / or one or more models (such as physics-based, statistical, and / or mathematical models) to generate new features and / or parameters. The combination and analysis of multiple machine functions and metadata may identify, for example, parameter problems that affect the machine, the process, the build, etc. For example, smoke and / or soot deposition may be detected, which indicates insufficient gas flow, incorrect speed / beam / power parameters (e.g., set too high, etc.), etc. The machine health and processing data together may indicate the impact on the build and part quality. The image data and / or log files may be stored in the server 210 , the user computing device 220 , and / or the mobile computing device 230 .

[0077] The server 210 typically includes a processor, memory, and chipset for communicating resources via the network 240. The resources may include, for example, processing, storage, software, and information provided from the server 210 to the user computing device 220 via the network 240. The server 210 may store and / or dynamically calculate machine learning models or statistical models of parameters / features from the additive manufacturing system 100. The user computing device 220 typically includes a processor, memory, and chipset for communicating data via the network 240.

[0078] refer to Figure 2 , the mobile computing device 230 may be any device having hardware (e.g., a chipset, a processor, a memory, etc.) for communicatively coupling with the network 240. Specifically, the mobile computing device 230 may include an antenna for communicating via one or more of the above-mentioned wireless computer networks. In addition, the mobile computing device 230 may include a mobile antenna for communicating with the network 240. Thus, the mobile antenna may be configured to send and receive data according to any generation (e.g., 1G, 2G, 3G, 4G, 5G, etc.) of mobile telecommunication standards. Specific examples of the mobile computing device 230 include, but are not limited to, smartphones, tablet devices, e-readers, laptop computers, etc. The mobile computing device 230 may have a display similar to the display device 408 of the user computing device 220, and display a user interface.

[0079] refer to Figure 2, the network 240 typically includes a plurality of base stations configured to receive and transmit data according to mobile telecommunication standards. The base stations are also configured to receive and transmit data through a wired system such as a public switched telephone network (PSTN) and a backhaul network. The network 240 may also include any network accessible via a backhaul network, such as a wide area network, a metropolitan area network, the Internet, a satellite network, etc. Therefore, the base station typically includes one or more antennas, transceivers, and processors that execute machine-readable instructions to exchange data through various wired and / or wireless networks.

[0080] Go to Figure 3 , showing Figure 1 1 and 2. In particular, the various internal components of the control unit (also referred to herein as the additive machine controller) 120 are depicted in FIG. Figure 3 Depicted is a method for collecting parameters and images to operate the additive manufacturing system 100, analyzing the parameter and image data, and / or assisting in controlling Figure 1 Various system components of the additive manufacturing system 100 depicted in FIG.

[0081] like Figure 3 As shown, the additive machine controller 120 may include one or more processing devices 302 , non-transitory memory components 304 , network interface hardware 308 , device interface hardware 310 , and data storage components 306 , all interconnected via a local interface 300 , such as a bus or the like.

[0082] One or more processing devices 302, such as a computer processing unit (CPU), may be a central processing unit of the additive machine controller 120 that performs computations and logic operations to execute programs. One or more processing devices 302, alone or in combination with other components, are illustrative processing devices, computing devices, processors, or combinations thereof. One or more processing devices 302 may include any processing component configured to receive and execute instructions, such as from data storage component 306 and / or memory component 304.

[0083] The memory component 304 may be constructed as a volatile and / or non-volatile computer readable medium, and thus may include random access memory (including SRAM, DRAM, and / or other types of random access memory), read-only memory (ROM), flash memory, registers, compact disks (CDs), digital versatile disks (DVDs), and / or other types of storage components. The memory component 304 may include one or more programming instructions thereon, which, when executed by the one or more processing devices 302, cause the one or more processing devices 302 to perform various processes.

[0084] Still refer to Figure 3The programming instructions stored on the memory component 304 may be embodied as a plurality of software logic modules, wherein each logic module provides programming instructions for completing one or more tasks.

[0085] Still refer to Figure 3 , network interface hardware 308 may include any wired or wireless network hardware, such as a modem, a LAN port, a wireless fidelity (Wi-Fi) card, a WiMax card, mobile communication hardware, and / or other hardware for communicating with other networks and / or devices. For example, network interface hardware 308 may be used to communicate with a wireless network via a wireless network such as a wireless router. Figure 2 The illustrated network 240 facilitates communications between the additive manufacturing system 100 and external devices such as the server 210 , the user computing device 220 , the mobile computing device 230 , and the like.

[0086] refer to Figure 3 , the device interface hardware 310 can be used between the local interface 300 and Figure 1 For example, the device interface hardware 310 may serve as a local interface 300 and a Figure 1 The device interface hardware 310 may be configured to interface between the imaging device 114, the powder distributor 108, etc. Figure 1 The imaging device 114 and / or other sensors transmit signals and / or data to or receive signals and / or data from it.

[0087] Still refer to Figure 3 , the data storage component 306 (which may generally be a storage medium) may include one or more data repositories for storing the received and / or generated data. The data storage component 306 may be any physical storage medium, including but not limited to a hard disk drive (HDD), a memory, a removable storage device, etc. Although the data storage component 306 is depicted as a local device, it should be understood that the data storage component 306 may be a remote storage device, such as a server computing device, a cloud-based storage device, etc. Illustrative data that may be included in the data storage component 306 includes but is not limited to image data 322, machine learning (ML) data 324, and / or operational data 326. Image data 322 may generally be used by the control component 120 to identify a specific object, determine the powder layer 112 ( Figure 1 ), monitor the amount of electromagnetic radiation at one or more points, determine changes in electromagnetic radiation, etc. For example, the additive machine controller 120 can access the image data 322 to obtain a plurality of images received from the imaging device 114, determine the amount of electromagnetic radiation based on the image data 322, and generate one or more commands accordingly.

[0088] Still reference Figure 3, the ML data 324 may be used as a tool for determining the powder layer 112 ( Figure 1 ) is the result of one or more machine learning processes or statistical modeling processes that can be used to identify the characteristics of a data set. Figure 3 , the operational data 326 may include parameters output from a plurality of subsystems of the additive manufacturing system 100. For example, the operational data 326 may include parameters output from a vacuum system, an inerting system, a beam system, a powder layering system, and the like. Specifically, the parameters for the beam system may include, but are not limited to, maximum supply voltage, minimum supply voltage, filament burn time, average preheat grid voltage, grid voltage drop after arc trip, average cathode power, average effective work function, average smoke count, smoke warning, average column pressure, number of arc trips, maximum deviation of grid voltage, grid voltage at 2 mA, and the like. The parameters for the vacuum system may include, but are not limited to, maximum chamber pressure, minimum chamber pressure, maximum column pressure, minimum column pressure, vacuum fault error, average change in chamber vacuum, minimum helium supply line pressure, average current of chamber turbo pump, average current of column turbo pump, turbo pump idle duration, average internal loop temperature, average inlet cooling water temperature, and the like.

[0089] It should be understood that Figure 3 The components shown in are illustrative only and are not intended to limit the scope of the present disclosure. More specifically, although Figure 3 The components in are shown as being located within the additive machine controller 120, but this is a non-limiting example. In some examples, one or more of the components may be located outside the additive machine controller 120.

[0090] Figure 4 Depicted Figure 2 The various internal components of the user computing device 220 are depicted in FIG. Figure 4 As shown, the user computing device 220 may include one or more processing devices 402, non-transitory memory components 404, network interface hardware 406, display devices 408, and data storage components 410, all of which are interconnected via a local interface 400 (such as a bus, etc.). Figure 4 Components of a user computing device 220 are depicted, but Figure 2 The server 210 in may have Figure 4 The same or similar parts are shown.

[0091] One or more processing devices 402, such as a computer processing unit (CPU), may be a central processing unit of the user computing device 220 that performs computations and logic operations to execute programs. One or more processing devices 402, alone or in combination with other components, are illustrative processing devices, computing devices, processors, or a combination thereof. One or more processing devices 402 may include any processing component configured to receive and execute instructions, such as from data storage component 410 and / or memory component 404.

[0092] The memory component 404 may be configured as a volatile and / or non-volatile computer readable medium, and thus may include random access memory (including SRAM, DRAM, and / or other types of random access memory), read-only memory (ROM), flash memory, registers, compact disks (CDs), digital versatile disks (DVDs), and / or other types of storage components. The memory component 404 may include one or more programming instructions thereon, which, when executed by the one or more processing devices 402, cause the one or more processing devices 402 to diagnose a component or a construct of an additive manufacturing system.

[0093] Still refer to Figure 4 , display device 408 may include any medium capable of transmitting light output, such as a cathode ray tube, a light emitting diode (LED) display, an organic light emitting diode (OLED) display, a liquid crystal display, a plasma display, etc. In some examples, display device 408 may be a touch screen that, in addition to visually displaying information, detects the presence and location of tactile input on or adjacent to the surface of display device 408 .

[0094] Still refer to Figure 4 , the data storage component 410 (which may generally be a storage medium) may include one or more data repositories for storing the received and / or generated data. The data storage component 410 may be any physical storage medium, including but not limited to a hard disk drive (HDD), a memory, a removable storage device, etc. Although the data storage component 410 is depicted as a local device, it should be understood that the data storage component 410 may be a remote storage device, such as a server computing device, a cloud-based storage device, etc. Figure 4 As shown in the example of , the data storage component 410 can store one or more models and / or data, such as a physical model 411, a data science model 412, parameter data 414, a diagnostic model 416, a hybrid model 418, etc.

[0095] Figure 5 An example infrastructure or framework 500 for monitoring and adjusting additive manufacturing based on machine health monitoring and analysis is shown. Figure 5In the example framework 500 of , the additive manufacturing machine 100 and its controller 120 communicate with an analysis processor 510, which uses the melt pool data acquisition 520 and information from the additive manufacturing machine 100 and / or its control computer 120 to model the additive machine 100 and its operation, assess the health of the additive manufacturing machine 100, score or otherwise assess the build quality (e.g., real-time or near real-time delamination, etc.), and / or otherwise monitor, model, assess, and modify the configuration, performance, control, etc. of the additive manufacturing machine 100. Although in Figure 5 In the example of FIG, analysis processor 510 is shown as a single device, but analysis processor 510 can be implemented as, for example, one or more FPGAs, processors, trusted platform modules (TPMs), and / or other computing devices. When analysis processor 510 is implemented using multiple devices, the FPGAs and processors can coordinate with each other to achieve common data fusion and / or analysis. For example, multiple devices can be synchronized to the same time base, build metadata, and / or machine control events.

[0096] Multiple processing parameters affect the microstructure and mechanical properties of 3D printed objects using powder bed fusion and / or other laser-based melting processes, including scanning speed (e.g., in millimeters per second (mm / s), etc.), beam velocity / velocity function, beam current or beam power (e.g., in watts (W), etc.), layer thickness (e.g., in mm, etc.), powder layer uniformity, and line offset. These parameters can be adjusted and / or optimized to obtain the desired 3D printed object properties. For example, beam power, scanning speed, spacing, layer thickness, etc. affect energy density (e.g., average applied energy / unit volume of material, J / mm3). In some examples, the beam speed can be adjusted near the edge of the object to prevent overheating. Layer thickness (e.g., 50-150um) affects the geometric accuracy of the object and can vary depending on the type of 3D printer used and other processing parameters (such as material powder particle size, etc.). In addition, the scanning pattern and scanning speed will also affect the final 3D printed object microstructure and porosity. For example, a scan pattern (e.g., a cross section of a layer) represents the geometric trajectory of an electron beam, laser beam, and / or other energy beam 131 used to melt metal powder to form a cross section on a powder bed / build area. For example, such a geometry may include an outer contour, an inner contour, and / or a shadow pattern.

[0097] Problems with other components besides the energy beam 131 may also lead to build defects and / or processing flaws. For example, obstacles or interruptions in obtaining and spreading powder on the build plate or platform 111 may also lead to defects in the resulting built parts. Other components can also be analyzed, and associated anomalies or problems can be identified / predicted and solved. For example, the laser / emitter 130 may have low power problems, material problems, etc. The scanner may experience problems such as wear (e.g., detectable via a pattern), speed (e.g., detectable relative to time). For example, wear can also be evaluated relative to movable components (such as pumps, motor build chambers, motor powder chambers, motor recoaters, building modules). For example, the accuracy of one or more sensors (e.g., optical temperature sensors, oxygen sensors, pressure sensors, temperature sensors, dew point sensors, etc.) can be evaluated. The function and / or performance of one or more components (such as collimators, optical coolers, valves, seals, etc.) can be evaluated. For example, the control board can be evaluated relative to operating time, mean time to failure (MTTF), etc.

[0098] The parameters / settings involved in other modes may be different from those described above. For example, in the BinderJet system, single-layer statistical process control (SPC) and multi-layer SPC can be facilitated. Such SPC can include image-based defect detection, including powder bed monitoring, with short dispersion and closed-loop control to trigger another dose and recoat. SPC can also include image-based defect detection with powder bed monitoring, where streaks / pushing of powder indicates that the recoater is damaged, resulting in recoating closed-loop control or stopping the build. Image-based defect detection can also include using pattern test image analysis in a closed loop (e.g., pre-printing, spraying patterns on paper, imaging and analysis, etc.) to automatically re-clean the jetting health of the jet with a solvent. In some examples, the jetting health can include analysis of the jetting onto the powder in a closed loop (e.g., potentially imaging the binder on the powder with an IR camera, and checking the binder deposited into the powder, etc.) to automatically re-clean the jetting with a solvent. Image-based defect detection can also include analysis of the geometry of the jetting onto the powder and the expected geometry. For example, images can be used to compare the expected geometry with the actual geometry. BinderJet analysis can also involve closed-loop control of curing parameters, recoating parameters, jetting parameters, etc. Curing parameters (e.g., IR lamp intensity, etc.) can be modified based on real-time (or substantially real-time) health conditions (e.g., based on powder bed temperature, images, etc.). Recoating parameters (e.g., dose coefficient, roller speed, etc.) can be modified based on real-time (or substantially real-time) health conditions (e.g., based on powder bed temperature, images, etc. from IR cameras). Jetting parameters (e.g., measuring / modifying saturation, manifold pressure, drop rate, etc.) can be modified based on real-time health metrics (jetting quality, saturation, etc.) to enable detection, correction, and preventive measures. Other parameters (such as humidity, temperature, pressure, drive / pump current, and torque) all affect bonding speed, quality, saturation of adhesive entering the powder bed, and overall part quality. These parameters can be combined to monitor the IR lamps for each layer of the curing build (e.g., higher humidity affects curing time and saturation). Lower IR temperatures caused by lamp degradation can affect print time and overall quality, etc.

[0099] In some examples, one sensor may provide information about one parameter for a subsystem. In other examples, multiple sensors may be assigned to a subsystem such that different types of sensors provide different types of data about one or more parameters of the subsystem. For example, using multiple sensors, more correlations may be established between the parameters of the subsystem to drive improved error prediction, detection, and correction.

[0100] Certain examples provide multimodal data fusion for near real-time detection, diagnosis, and prediction of additive machine and process health and build quality. The layered printing process of additive manufacturing technology allows data of the printing process to be captured from the first layer to the last layer of the printed part. This data can include, but is not limited to, part design, material parameter definitions, machine settings and configurations, as well as sensor and programmable logic controller (PLC) data, log information, software errors / warnings, and other build information such as recoat imaging and print videos.

[0101] For example, within a given build, intelligent data weighting can be used to identify anomalies on one or more layers. Attenuation strategies can also be employed on one or more layers, as a single layer anomaly can sometimes become a multi-layer problem. However, other times, a problem in one layer can be naturally cured by thermal effects from surrounding layers, etc. For example, weighting factors and employing attenuation can help identify anomalies and assess whether the anomaly is likely to spread to other layers or be corrected by other layers.

[0102] Certain examples provide a base data architecture to facilitate data fusion and establish a complete digital lineage for each 3D print. For example, on this basis, statistical process control methods can be used to monitor machine, process, and part health in situ throughout the build process on a hierarchical basis to detect anomalies, defects, and enable closed-loop control and correction. In certain examples, the same machine and / or multiple machines may experience variability across multiple builds. For example, one or more additive manufacturing machines that repeatedly build the same part can utilize this base data to compare / contrast build health, including statistical processor control analysis of subsystem components and process variables / configurations.

[0103] Certain examples enable real-time monitoring of additive build processes, machine quality, and build quality. During the build, as layers are created, health and performance analysis is performed to determine if there are any critical issues so that the operator can take timely action. The health and performance analysis may involve build and / or print health, machine performance, and / or part quality, etc. Multimodal data (e.g., time series data from installed sensors, simulation / compensated modeling results, build inputs, settings, images from optical and / or infrared cameras associated with the powder bed, melt pool, etc.) is fed to a hybrid model that includes a series of analytical modules (e.g., data science, statistics, and additive domain-based physics). The hybrid model filters and / or otherwise pre-processes the data for feature extraction, signal-to-noise ratio (SNR) enhancement, etc., and then feeds the processed data to an analytical model that calculates physics-based metrics, which are further analyzed using data science models based on one or more methods (such as statistical process control (SPC), statistics, Bayesian, machine learning, etc.) to determine the overall quality of the process and the health of the additive manufacturing machine. The health status metrics and model results are further combined using a probabilistic model to determine an overall severity (e.g., health status) score for the layer and an individual quality score for each "critical x" associated with each modality. Single-layer scores, multi-layer scores, etc. can be determined and used to detect, correct, and prevent problems with the build, process, and / or device. For example, modalities can include, but are not limited to, DMLM, EBM, DCAT, and BinderJet.

[0104] The analysis enables health detection and diagnosis of the build process, AM machine, and build quality at both individual and multiple levels using multi-layer analysis, trend analysis, change point detection, cumulative damage assessment, etc. Cumulative damage assessment is performed while the build is ongoing based on the performance of previous layers to increase confidence in health and / or anomaly detection and correction as the build progresses, as well as overall health diagnosis of the entire build at the end of the build. In some examples, forecasts and / or other predictions can be provided based on previous build data, trend analysis, etc. to identify problems before they occur and / or before they become issues that affect the build of the part or machine performance.

[0105] These analyses are performed on a separate, secure computing device isolated from the control operations to minimize risks to the performance of the additive manufacturing machine. For example, the authenticity and integrity of the analysis application code is securely protected by one or more integrated trusted platform modules (e.g., implemented on one or more FPGAs, etc.). Thus, a root of trust is established between the analysis processor 510 and the additive manufacturing machine 100 (and in some examples, between the analysis processor 510 and the additive machine controller 120). The analysis results are updated in real time on a network-based human-machine interface (HMI). For example, the HMI can be accessed by the printer's control computer and / or remotely via a network connection.

[0106] like Figure 5 As shown in the example of , the analysis processor 510 provides live monitoring of the additive manufacturing machine 100 during the build of a part or a group of parts. The example analysis processor 510 can apply analysis to its real-time (or substantially real-time given transmission and / or processing delays, etc.) monitoring to determine the health of the additive machine 100. The analysis processor 510 receives and processes data during the build and provides automatic quality scores for the machine, process and / or build data, rather than extracting data from the machine 100 at the end of the build. For example, the analysis processor 510 can facilitate hierarchical process monitoring of machine health, build process health, and / or part quality. Such monitoring can be facilitated by collecting real-time (or substantially real-time) data from one or more on-machine sensors, cameras, etc. For example, the acquired information is analyzed using statistical and / or artificial intelligence (AI) models to determine quality scores for machine health, process health, and / or part quality. Data is transmitted to the analysis processor 510 from the additive machine 100, the additive machine controller 120, etc. In some examples, melt pool data from the additive machine 100 is collected by the melt pool data acquisition circuit 520 and provided to the analysis processor 510.

[0107] By providing the analysis processor 510 separately from the additive manufacturing machine 100 or the additive machine controller 120, the dedicated analysis processor 510 can perform high computational processing without reducing the efficiency or speed of the additive manufacturing machine 100 itself. The analysis processor 510 is isolated from the additive manufacturing machine 100 and processes information from the additive machine 100 without interfering with the operation of the additive machine 100 unless the processing reveals a problem and adjustments are to be made to the machine settings, process configuration, current and / or subsequent builds, etc. The analysis processor 510 can examine the specific processes of the additive manufacturing machine 100 (e.g., melting, recoating, etc.) on a layer-by-layer basis, as well as the overall health of the machine, process, build, etc. Certain examples provide a modular architecture that can be constructed and extended to multiple product lines (e.g., M2, Mline, etc.), modalities (e.g., EBM, binder jetting, etc.), etc.

[0108] The interaction of physics and parameters is built into the analysis of the analysis processor 510. For example, the knowledge of the interaction between the additive manufacturing machine 100, sensors, and material parameters and machine operation enriches the analysis. Correlations and physical properties may be affected by input parameters of the material, additive manufacturing machine 100 configuration (e.g., gas flow, beam velocity, and laser power, layer thickness of deposited powder, etc.), etc. For example, correlations may include log / time series, static images (e.g., thermal, still, infrared (IR), etc.), video, SPC and quality score metadata, and / or a fusion thereof.

[0109] Figure 5 An example infrastructure or framework 500 for monitoring and adjusting additive manufacturing based on machine health monitoring and analysis is shown. Figure 5 In the example framework 500, the additive manufacturing machine 100 and its controller 120 communicate with the analysis processor 510, and the analysis processor 510 uses the melt pool data acquisition 520 and information from the additive manufacturing machine 100 and / or its control computer 120 (for example, from the camera or other imaging device 114, the transmitter 130, the build plate / platform 111 and the positioner and / or other lifting components 113, the sensors 150-151, etc.) to model the additive manufacturing machine 100 and its operation, evaluate the health of the additive manufacturing machine 100, score or otherwise evaluate the build quality (for example, real-time or near real-time delamination, etc.), and / or otherwise monitor, model, evaluate and modify the structure, performance, control, etc. of the additive manufacturing machine 100.

[0110] Thus, the additive machine controller 120 may adjust or correct the operation of the additive machine 100 based on the information provided by the analysis processor 510. As described further below, the analysis processor 510 performs layered analysis to detect and / or predict errors or other problems, thereby driving corrections or modifications to affect the current layer, future layers, future builds, etc. For example, single-layer and / or multi-layer analysis by the analysis processor 510 may correct, optimize, and / or otherwise improve future layers.

[0111] For example, multi-layer laser health monitoring may be implemented by an analysis processor 510 that combines FPGA data capture and analysis of delivered laser power versus expected or commanded laser power. Over time, laser power may degrade and / or laser calibration may decay and / or drift. For example, the analysis processor 510 may monitor commanded power and position alone or in combination with one or more FPGAs and detect / correct alignment issues. Similarly, the health of the scanning galvanometer may also be observed across multiple layers, including drift in accuracy over time related to galvanometer temperature. The higher the temperature, the greater the likelihood that the scanner will experience wear, which can affect accuracy, etc.

[0112] For example, the analysis processor 510 provides analysis of the subsystems of the additive machine 100 and their processes to generate build process optimization, inspection recommendations, and / or predictive machine maintenance. The analysis processor 510 uses multi-sensor time series analysis, image analysis, correlation analysis, etc. across sensors, log files, and other machine health and processing data (images, melt pool emissivity, etc.). For example, the near real-time analysis of the analysis processor 510 enables the additive manufacturing machine 100 and / or associated users to make timely decisions, adjustments, etc. to affect the machine / materials and post-processing work-in-process (WIP) cost avoidance during the build (without waiting or relying on post-processing inspections). The analysis processor 510 uses automatic data transmission and data analysis for each layer as the build is printed. For example, the analysis processor 510 is an independent and secure analysis computing platform with a delay of no more than one layer behind the current printing layer, for example, which does not interfere with the additive machine 100, but allows for near real-time monitoring and adjustments to the additive manufacturing machine 100. For example, the analysis processor 510 may include a physics-based model that models the structure, construction, and operation of the additive machine 100 and its associated processing and construction using system expertise and additive physics. Alternatively or additionally, the analysis processor 510 may include one or more data science models for diagnosing and / or predicting machine, processing, and / or construction errors. In some examples, the analysis processor 510 uses one or more fusion models that combine physics and data science to determine / predict results. In some examples, the analysis processor 510 is constructed with a modular architecture to help ensure faster adoption of new product lines and modalities (e.g., EBM, DMLM, BinderJet, DCAT, etc.) by easily reusing and "exchanging" modality and / or product line software Docker containers (e.g., 20-30%) while retaining common analysis and software functions (e.g., 70%).

[0113] In some examples, the analysis processor 510 provides statistical process control (SPC) using automatic SPC calculations of the X "most important" variables for a given modality. The layer-by-layer analysis and visualization includes subsystems, processes, and subprocesses of the additive manufacturing machine 100. Multi-layer analysis is used by the analysis processor 510 to form a cumulative damage assessment as the build progresses and the overall end-of-build health state. Instead of or in addition to multi-layer analysis, a large number or various sensors can be analyzed with SPC to form a cumulative assessment. For example, SPC can be used to calculate individual sensor values, which can then be combined on a layer for hierarchical scoring. Therefore, one or more data sets for each sensor and / or one or more sensors can be evaluated, combined, etc. For example, the analysis processor 510 can use a series of analyses activated based on modality (e.g., DMLM, EBM, BinderJet, DCAT, etc.), target (which component or process), etc. to achieve machine and / or modality agnosticism.

[0114] In some examples, hybrid models (e.g., hybrid AI models) provide a fusion of additive process physics, machine know-how, and data science (e.g., SPC / statistics / Bayesian / machine learning) to process data from the additive machine 100 and / or the melt pool analyzer 520. "Default" and "configurable" analysis modules work with the model and implement machine and / or modality agnosticism in the analysis processor 510. For example, a default module may include circuits and instructions for calculating and extracting statistical features. Examples of configurable modules (e.g., physics-based modules) are circuits and instructions for calculating "filter clogging coefficient" (for DMLM), "cathode health coefficient", "print head health" (for BinderJet), "light projector health" (for DCAT), etc. For example, a configurable analysis can change the sensitivity of the detection. For example, a "sensitivity slider" can set an increase or decrease in sensitivity for test / benchmark vs. "production / qualified" builds. Configurable software modules allow for the construction and selection of data types, data frequencies, storage locations, etc. (e.g., where the data includes sensor data, log data, image data, build parameters, etc.). For example, within a build, layered monitoring of machine and process health performance through sensors, melt pool data, powder bed information, etc., enables automated inspection and modification of the machine, process, and / or build. The analysis processor 510 provides a dedicated computing device to help ensure that computing resources are isolated from the additive machine controller 120 and the control operation of the additive machine 100. The analysis sensitivity and / or severity of reported alarms can be configured based on defined areas of interest within the build. For example, image-based analysis is configured based on area definitions, which can include but are not limited to part boundaries, layer ranges, bounded regions, or bounded volumes.

[0115] In some examples, the authenticity and integrity of the analysis code is protected with one or more integrated trusted platform modules (TPMs). The use of a TPM can provide a hardware root of trust (e.g., establish a trust relationship) for the analysis processor 510, the additive machine controller 120, etc. For example, a TPM can provide a tamper-proof ground for detecting and correcting errors. For example, a root of trust can include ensuring the authenticity of a signed and / or otherwise authenticated analysis. In some examples, a combination of a TPM and code signing provides enhanced security with a root of trust to enable the analysis processor 510 to communicate with and affect the machine controller 120, the additive machine 100, etc. For example, a TPM supports disk encryption and a boot chain to protect the authenticity and integrity of the code on the analysis processor 510. Signed / authenticated packages can be sent from the analysis processor 510 to the additive machine controller 120, and / or directly to the additive manufacturing machine 100 for security updates.

[0116] Instead of, or in addition to, physically based models, data science models, and / or hybrid models, neural network (NN) or convolutional neural network (CNN) models, random forests, and other AI / machine learning (ML) methods involving big data, artificial defect characterization, and model training can be used to provide data-driven analysis. In some examples, the machine learning model is trained in an offline system that has access to a large data set, and the trained model is then deployed to the analysis processor 510. In some examples, the training model is assisted by the use of a computing accelerator (such as a GPU, VPU, and / or FPGA). Therefore, the model can be trained on one or more systems / builds, etc. based on previous build data (e.g., showing "good" builds, errors, corrections, etc.), "golden" reference build data, etc.

[0117] Thus, certain examples provide real-time or substantially real-time layered processing and machine health monitoring. A layered view of process variable build "vital signs" with SPC metrics is provided using the analysis processor 510 to timely identify process and machine performance deviations (e.g., delamination during build, etc.). A multi-layered process view can also be provided for one or more geometrically complex areas of interest of the build. Multi-layer analysis provides more insights into specific zones / areas with more complex geometries, while simpler areas of the part may only involve a single layer of analysis. The analysis processor 510 is capable of performing a mix of single and multi-layer analysis depending on part geometry, construction, monitoring criteria, etc. For example, the analysis processor 510 provides a "one-stop" solution for process monitoring features, sensors, and powder bed / binder applications.

[0118] In some examples, machine health and build quality monitoring is integrated with automatic data transmission (e.g., via interface 530) from the additive machine 100, controller 120, and / or melt pool data analyzer 520. For example, sensors on the machine, data transmission, and single layer analysis can provide layer health scores as well as SPC metrics. In some examples, the analysis processor 510 can use multi-layer sensor analysis to determine trends, detect change points, etc. The analysis of the analysis processor 510 can be integrated with modality-specific closed-loop control (including recoat monitoring of the additive machine 100, adhesive application monitoring, powder bed monitoring (e.g., for short feeds, protrusions, etc., in combination with gas flow, oxygen content, etc.), etc.). In some examples, the monitored layer can be compared to a reference (e.g., "golden") layer to identify errors and / or other differences in the current build. For example, the reference or "golden" layer can be digitally determined and / or generated from the actual build. For example, CT and / or other imaging can be used to inspect the build and / or build layer and verify its quality and acceptability to be designed as a golden / reference layer / build. In some examples, a reference or golden build formed from multiple layers may be utilized.Overall Equipment Effectiveness (OEE), additive manufacturing machine 100 efficiency, and overall build health quality scores may be determined (eg, hierarchical and multi-layer analysis).

[0119] Thus, the example system 500 may be used to monitor, assess, and address machine health. Machine failures may be detected and isolated by an analysis processor 510 (e.g., implemented as one or more processors, FPGAs, other computing devices, etc., to detect and / or predict additive manufacturing machine 100 failure modes based on field data). Machine health issues may include degradation of the optics, recoaters, gas flow, laser glass contamination detection, etc. Automatic calibration of the optics (e.g., beam alignment, multi-laser alignment, closed-loop spot size control, power calibration, etc.) may also be driven by machine health monitoring and analysis.

[0120] As described above, build and / or AM machine health / issues, successes / failures, etc. may be monitored and evaluated. Such analysis may be extended to one or more builds on one or more machines to leverage problem identification in one build on one machine and infer causal relationships and impact on resolution for one or more future builds on one or more AM machines. Figure 6 An additive manufacturing machine behavior apparatus or infrastructure 600 is shown. The example apparatus or infrastructure 600 includes a plurality of additive manufacturing machines 100, 602, a feature engine 610, a learner circuit 620, an evaluator circuit 630, and a data storage 640.

[0121] For example, the data storage device 640 stores one or more log files, sensor data, images, etc. captured from the additive manufacturing machine 100, 602. Figure 6 As shown in the example of , data stored in the data storage device 610 is ingested by the feature engine 610 to identify features from the stored data. For example, layer features, build features, etc. can be extracted and aggregated for further analysis. Features can be sensor features (e.g., temperature, gas flow, etc.), machine setting features (e.g., dose factor, recoating, laser shots, etc.), build features (e.g., total build time, exposure time, recoating time, etc.), etc.

[0122] The example learner circuit 620 processes the log file data and associated features to learn machine behavior using one or more artificial intelligence models (e.g., neural network models, other machine learning models, random forest models, etc.) from the build population represented in the data. The learner circuit 620 learns to distinguish between standard or normal or acceptable builds and non-standard or abnormal or unacceptable builds based on analysis of the available data. The learner circuit 620 also determines which parameters / settings / conditions contribute to non-standard builds, which contribute to standard builds, etc. The learner circuit 620 can then train a model based on the build population represented in the data and the associated learning behaviors, relevant factors, and results. The trained model is then deployed as (part of) the evaluator circuit 630 to evaluate and identify non-standard (relative to standard) builds.

[0123] The example evaluator circuit 630 utilizes the deployed training model to evaluate the new build according to the saved learning behavior of the population represented by the training model to evaluate whether the new build is non-standard. The evaluator circuit 630 uses the training model to quantify the new build behavior and determine which additive manufacturing machine parameters and / or build settings cause such behavior. In some examples, the learner circuit 620 is centralized to benefit from training on various builds and additive manufacturing machine data, and the evaluator circuit 630 is a distributed plurality of evaluator circuits 530 associated with individual additive manufacturing machines 100, 602, etc., and / or associated analysis processors 510 and / or additive machine controllers 120. The evaluator circuit 630 can diagnose non-standard builds and communicate directly with the additive machine 100, 602 and / or communicate indirectly with the additive machine 100, 602 through its corresponding analysis processor 510 and / or controller 120 to adjust the additive manufacturing machine 100, 602 and / or its build to transform the behavior into standard behavior (e.g., for successful builds).

[0124] Figure 7-8 Additional views of an example additive manufacturing machine behavior infrastructure 600 are provided. Figure 7In the example of , the additive manufacturing machine 100, 602 provides input to the feature engine 610, and the learner circuit 620 receives data from the machine 100, 602 processed by the feature engine 610, and a list of features 720 with associated known limits (e.g., pump health, filter life, etc.). For example, the features and associated limits 720 may include the number of remelts in a layer (e.g., less than two, etc.), the number of consecutive layers with remelts (e.g., less than three, etc.), the average idle time per layer (e.g., less than eleven seconds, etc.), the number of arc trips within a ten-minute rolling duration (e.g., less than three, etc.), smoke flags (such as "zero pulse rake" counts after an arc trip) (e.g., less than five, etc.), the drift of cathode resistance during the build (e.g., less than 0.03 ohms, etc.), the number of rake stuck events (e.g., less than one, etc.), the average chamber vacuum pressure (e.g., less than 2.5e-3 mbar, etc.).

[0125] For example, the learner circuit 620 trains and / or utilizes one or more multivariate models 730 and one or more univariate models 735. The multivariate model 730 processes a combination of features to provide a relative analysis with a build of unsupervised learning (e.g., a combination of k-nearest neighbor (k-NN) anomaly detection with HDBScan clustering, etc.). The univariate model 735 evaluates one feature at a time. For example, the univariate model 735 can fit a feature to an optimal statistical distribution and extract relevant limits associated with the behavior of the feature value. For example, non-compliance with acceptable limits for a feature can be quantified as a non-compliance severity metric.

[0126] like Figure 7 As shown in the example of , the outputs of the multivariate model 730 and the univariate model 735 are provided to a data store 740 (e.g., the same, similar, or different from the example data store 640), and output 750 is as build-side results, such as builds marked as standard or non-standard, non-compliance severity scores, prominent features, machine-side spreads of standard and non-standard results, etc. The output 750 may also include feature-side results, such as best fit distribution and association plots, statistics for each feature (e.g., center, dispersion, quartiles, etc.), feature rankings based on contribution to non-standard builds, lists of abnormal builds, etc. Feature examples may include the number of remelts in a layer, the number of consecutive layers with remelts, the average idle time per layer, the number of arc trips within a ten-minute rolling duration, smoke flags (e.g., "zero pulse rake" counts after arc trips), drift of cathode resistance during a build, the number of rake stuck events, average chamber vacuum pressure, etc. The output 750 may also include layering information, such as median profiles of key features, layer level plots, etc. Key layering features may include preheat time per layer, melt time per layer, etc. Additional outputs 760 include anomaly thresholds, feature limits, etc.

[0127] The outputs 750, 760 may be stored in the data storage device 740, provided to another system, displayed, provided to the evaluator circuit 630, etc. The evaluator circuit 630 processes the features extracted from the additive manufacturing machine 710 by the feature engine 612 (e.g., the same or similar to the feature engine 610). The evaluator circuit 630 utilizes information from the data storage device 740 and / or the deployment model to cross-check thresholds, limits, etc. associated with the features. The evaluator circuit 630 may then generate an overall assessment of the build, a layer map with exceptions, and a non-compliance severity metric or score at the build level (e.g., for the overall build), feature level (e.g., for a specific feature of the build, a feature of a layer of the build, etc.), layer level (e.g., for a specific layer of the build), etc. The build may then be identified as standard or non-standard, and associated features that contribute to the classification may be determined, adjusted, used as feedback / reference, etc.

[0128] Figure 8 Shows Figure 7 , where the learning circuit 620 utilizes offline expert and / or automatic online learning for multiple build / machine populations based on features extracted from the additive manufacturing machines 100, 602 by the feature engine 610. The learning output and associated models can be stored in a data storage device 740 and provided as an output 750, such as an interactive dashboard that allows the selection of a population to view associated behavior, use the behavior as a reference and / or feedback, adjust the associated build / machine, etc. In addition, the evaluator circuit 630 pulls from the reference population behavior of the data storage device 740 to process the features of the new build extracted from the additive manufacturing machine 710 by the feature engine 612. The evaluator circuit 630 outputs 770 a comparison of the behavior of the new build with the behavior of the population or reference group of builds to evaluate whether the build is standard or non-standard. The evaluator circuit 630 also generates an explanation including a total non-compliance severity score, the number of non-compliant features, the highest / most contributing features, etc.

[0129] For example, an evaluation build in which a beam component of an additive manufacturing machine 710 has shown abnormal operating behavior may indicate that the number of arc trip events and / or drift in cathode resistance are major contributors to build noncompliance. In an example, an evaluation build in which a beam component of an additive manufacturing machine 710 has shown abnormal operating behavior may indicate that characteristics such as average idle time and / or rake jam events are significant contributors to build noncompliance.

[0130] Thus, the learner circuit 620 processes the build population and the additive manufacturing machine data to train multiple models (e.g., multivariate models and univariate models) to form a deployable build behavior model structure with the evaluator circuit 630 to determine whether a build is a standard build or a non-standard build. Fig. 9As shown in the example of , a build population 910 is provided to a learner circuit 620. The learner circuit 620 analyzes the population 910, learns patterns and correlations in the population data, and derives standard build behaviors 920 from the population 910. The learner circuit 620 can output a build population 930 with labeled behaviors. For example, the labeled population output 930 identifies a standard build, a non-standard build, etc., and can be used by the evaluator circuit 630 on new build data to determine whether the build is a standard build, a non-standard build, and why (e.g., which factor contributes most to making the build non-standard, etc.).

[0131] Although Figure 1-9 , an example implementation of the additive machine 100, additive machine controller 120, analysis processor 510, melt pool data acquisition processor 520, feature engine 610, learner circuit 620, evaluator circuit 630, data storage device 640, etc. is shown, but one or more of the elements, processes and / or devices may be combined, divided, rearranged, omitted, eliminated and / or implemented in any other manner. In addition, Figure 1-9 One or more of the elements of may be implemented by hardware, software, firmware, and / or any combination of hardware, software, and / or firmware. Thus, for example, Figure 1-9 Any of the example elements of may be implemented by one or more analog or digital circuits, logic circuits, programmable processors, programmable controllers, graphics processing units (GPUs), video processing units (VPUs), accelerator cards, digital signal processors (DSPs), application specific integrated circuits (ASICs), programmable logic devices (PLDs), and / or field programmable logic devices (FPLDs). When any device or system claim of this patent is read to cover pure software and / or firmware implementations, Figure 1-7 At least one of the example elements of is expressly defined herein as comprising a non-transitory computer-readable storage device or storage disk, such as a memory, a digital versatile disk (DVD), a compact disk (CD), a Blu-ray disk, etc., including software and / or firmware. Furthermore, in addition to Figure 1-9 In addition to or in place of those shown Figure 1-9 Those shown in Figure 1-9 The elements may include one or more elements, processes and / or devices, and / or may include more than one of any or all of the elements, processes and devices shown. As used herein, the phrase "communication", including variations thereof, encompasses direct communication and / or indirect communication through one or more intermediate components, and does not require direct physical (e.g., wired) communication and / or continuous communication, but additionally includes selective communication at periodic intervals, predetermined intervals, non-periodic intervals and / or one-time events.

[0132] Fig.10600. A flowchart representing example hardware logic, machine-readable instructions, hardware-implemented state machines, and / or any combination thereof for implementing all or part of an example additive manufacturing machine behavior apparatus or infrastructure 600 is shown in FIG. The machine-readable instructions may be instructions for executing by a computer processor and / or processor circuit (e.g., as described below in conjunction with Fig.13 The example learner circuit 620, the example evaluator circuit 630, and / or the example processor platform 1300 discussed above may be executed by one or more executable programs or portions of executable programs. The programs may be embodied in software stored on a non-transitory computer-readable storage medium such as a CD-ROM, floppy disk, hard drive, DVD, Blu-ray disk, or memory associated with the processor 1312, but the entire program and / or portions thereof may alternatively be executed by a device other than the processor 1312 and / or embodied in firmware or dedicated hardware. In addition, although reference is made to Figure 10-12 The flowchart shown in describes an example program, but many other methods of implementing all or part of the example additive manufacturing machine behavior device or infrastructure 600 may be used instead. For example, the order of execution of the boxes can be changed, and / or some of the boxes described can be changed, eliminated, or combined. Additionally or alternatively, any or all of the boxes can be implemented by one or more hardware circuits (e.g., discrete and / or integrated analog and / or digital circuits, FPGAs, ASICs, comparators, operational amplifiers (op-amps), logic circuits, etc.) that are constructed to perform corresponding operations without executing software or firmware. The processor circuits can be distributed in different network locations and / or located locally in one or more devices (e.g., a multi-core processor in a single machine, multiple processors distributed across server racks, etc.).

[0133] The machine-readable instructions described herein may be stored in one or more of a compressed format, an encrypted format, a segmented format, a compiled format, an executable format, a packaged format, etc. The machine-readable instructions as described herein may be stored as data or data structures (e.g., stored as portions of instructions, codes, representations of codes, etc.) that can be used to create, manufacture, and / or generate machine-executable instructions. For example, the machine-readable instructions may be segmented and stored on one or more storage devices and / or computing devices (e.g., servers) located in the same or different locations (e.g., in the cloud, edge devices, etc.) of a network or a collection of networks. The machine-readable instructions may need to be installed, modified, adapted, updated, combined, supplemented, constructed, decrypted, decompressed, unpacked, distributed, redistributed, compiled, etc., so that they can be directly read, interpreted, and / or executed by a computing device and / or other machine. For example, the machine-readable instructions may be stored in multiple parts that are individually compressed, encrypted, and stored on separate computing devices, where the parts form a set of executable instructions when decrypted, decompressed, and combined, and the set of executable instructions implements one or more functions that can together form a program such as described herein.

[0134] In another example, machine-readable instructions may be stored in a state where they can be read by processor circuitry, but require the addition of a library (e.g., a dynamic link library (DLL)), a software development kit (SDK), an application programming interface (API), etc., in order to execute the instructions on a particular computing device or other device. In another example, it may be necessary to construct the machine-readable instructions (e.g., stored settings, data inputs, recorded network addresses, etc.) before the machine-readable instructions and / or corresponding programs can be executed in whole or in part. Therefore, as used herein, a machine-readable medium may include machine-readable instructions and / or programs regardless of the particular format or state in which the machine-readable instructions and / or programs are stored or otherwise at rest or in transmission.

[0135] The machine-readable instructions described herein may be represented by any past, present, or future instruction language, scripting language, programming language, etc. For example, the machine-readable instructions may be represented using any of the following languages: C, C++, Java, C#, Perl, Python, JavaScript, Hypertext Markup Language (HTML), Structured Query Language (SQL), Swift, etc.

[0136] As mentioned above, Figure 10-12The example processes of can be implemented using executable instructions (e.g., computer and / or machine readable instructions) stored on a non-transitory computer and / or machine readable medium (such as a hard drive, flash memory, read-only memory, compact disk, digital versatile disk, cache, random access memory, and / or any other storage device or storage disk in which information is stored for any duration (e.g., an extended period of time, permanently, a transient instance, temporarily buffered and / or cached information). As used herein, the term non-transitory computer readable medium is expressly defined to include any type of computer readable storage device and / or storage disk and to exclude propagating signals and to exclude transmission media.

[0137] Fig.10 1 is a flow chart of an example method 1000 for monitoring, determining, and adjusting the health of a machine, process, and / or build, such as using an example learner circuit 620 and an example evaluator circuit 630. At block 1010, data of a population of builds on one or more additive manufacturing machines 100, 602 is processed. For example, the build data is ingested and formed according to identified features, formats, time series organization, etc. At block 1020, behavior is learned from the processed data. For example, data of multiple builds (e.g., multiple builds at a single site, across multiple sites, at the same machine, across multiple machines, etc.) is analyzed to learn which builds behaved as expected, which builds behaved unexpectedly, which builds had specific problems, which builds had specific solutions, which builds had other results, etc.

[0138] For example, the learner circuit 620 reads the log files for a given population of builds and learns the machine behavior associated with these builds by performing a relative analysis of just these populations of builds using unsupervised machine learning and statistical methods. This enables finding non-standard builds in the population and key machine parameters that lead to such behavior.

[0139] At block 1030, reference standards and non-standard behaviors are generated. For example, based on the builds and machine behaviors learned by the example learner circuit 620, the learner circuit 620 can generate standard reference behaviors for successful or other normal builds, and non-standard reference behaviors for unsuccessful builds. For example, the standard reference behaviors and non-standard reference behaviors can be used to compare other builds. Therefore, the standard reference behaviors and non-standard reference behaviors can be stored in digital form for use by the evaluator circuit 630.

[0140] At block 1040, a composite model is generated to evaluate and classify the build. For example, the composite model is trained by standard reference behaviors and non-standard reference behaviors. By training and testing the model using known / validated standard and non-standard behaviors, the model can be used to classify new builds as standard or non-standard based on the quantified behaviors reflected in the model. In some examples, machine parameters that cause such behavior can also be identified. For example, identifying machine parameters and / or parameter values ​​may be important to correct non-standard build / machine behavior.

[0141] Thus, the composite model can identify and explain variation at the build level, system level, subsystem level (e.g., bundle subsystem, etc.), etc. The composite model can be deployed on various populations (e.g., current build, builds in the past six months, builds in the past year, etc.). The composite model can be trained and tested against various AM machine populations (e.g., DMLM, EBM, BinderJet, etc.).

[0142] At block 1050, the composite model is deployed with the example evaluator circuit 630. For example, the composite model is deployed as part of the example evaluator circuit 630 to evaluate one or more ongoing builds, previous builds, etc. The composite model and / or associated reference standard and non-standard build behaviors may also be stored in a reference database (such as a data storage device 640, 740) for access by the evaluator circuit 630. At block 1060, new build data is processed to identify whether the build is standard or non-standard. For example, the evaluator circuit 630 receives build data from one or more of the additive machines 100, 602, the data storage device 640, the feature engine 610, etc. The evaluator circuit 630 provides data to the composite model to determine whether the build is non-standard or compliant with the standard based on the modeled behavior. For example, by applying learned, modeled thresholds, limits, etc. to the build, one or more features and / or the overall build may be evaluated to identify current issues, predict future issues, etc., and associated severity.

[0143] At block 1070, for a non-standard build, factors that contribute to the build being non-standard are identified. For example, when a build is identified as non-standard compared to the learned non-standard build behavior of the composite model, one or more factors, features, etc. that appear to contribute to the non-standard behavior of the build are identified (e.g., pre-heating, post-heating, melting time, etc.). Identifying contributing factors can be used to drive solutions to the non-standard build behavior (e.g., for a specific machine used for the build, other machines of the same type, other builds, etc.).

[0144] At block 1080, a designation is output for the build. For example, the evaluator circuit 630 outputs a designation or identification that the build is a standard or conforming build (e.g., a build that meets or complies with its build, quality metrics, etc.), or that the build is a non-standard or non-conforming build (e.g., a build that does not meet its build, does not meet associated quality metrics, has failed, etc.).

[0145] At box 1090, when the build is identified as non-standard, a corrective action is output. For example, based on the identification of one or more contributing factors / features driving the non-standard behavior, an associated solution can be generated. In some examples, the corrective instructions are recorded and transmitted to the operator for adjustment. In other examples, the additive machine 100, 602, the additive machine controller 120, etc. execute the corrective instructions, thereby adjusting the machine to correct the build. In some cases, the non-standard build is considered a failure and scrapped. In other cases, the non-standard build can be saved and / or otherwise corrected by executing the corrective instructions during the process. For example, active maintenance, automatic changes in configuration / settings, etc. can be triggered based on the detection of abnormal (e.g., non-standard) behavior in the build before the part goes bad.

[0146] At block 1095, available feedback, if any, is evaluated to improve the model. For example, feedback from the additive machine 100, 602, the additive machine controller 120, etc. may be captured and provided to the learner circuit 620 to adjust the composite model.

[0147] Fig.11 Further example details regarding learning and model building for the example learner circuit 620 described above are provided. At block 1110, the learner circuit 620 ingests data from a build population. For example, log files, sensor data, other inputs from one or more additive machines 100, 602, additive machine controllers 120, etc. are collected for processing by the learner circuit 620. In some examples, the feature engine 610 processes log files from a build population (e.g., collected over a year, several months, etc.), the feature engine 610 processes the log files to extract information, converts the information into features, and merges the features (e.g., problems in layers, vacuum, pre-heating, post-heating, etc.) into multivariate models and univariate models for the learner circuit 620 (e.g., merging data from hundreds of builds into two sets of models for the learner circuit 620, etc.). Models can be formed based on identified features, formats, time series organization, etc.

[0148] Multivariate models and univariate models are complementary and can be used together. Multivariate models are employed to capture nonlinear relationships in a multidimensional space containing various parameters, thereby identifying areas of standard behavior in the high-dimensional space. Univariate models provide an easily interpretable way to enable identification of the main contributors to nonstandard build behavior, and the system can then take action to correct it. For example, for evaluator building, the non-compliance score for each parameter is assessed individually based on its incidence relative to limits inferred from known standard behavior, and the scores are then combined in a linear weighted approach.

[0149] At block 1120, the learner circuit 620 learns build behaviors from the ingested data. For example, the learner circuit 620 processes the data (e.g., processes merged features formed from the data, etc.) to learn build behaviors by evaluating and classifying build behaviors. These features may be analyzed individually and / or in combination to organize and learn build behaviors from the population of data available to the learner circuit 620.

[0150] At block 1130, the learner circuit 620 generates a model and / or other analysis based on the learned build behavior. For example, a multivariate model can be generated to evaluate multiple features relative to each other and mark a build as standard or non-standard based on the comparison. A univariate model looks at individual features that have normal values ​​or ranges, such as melting time, etc., and can compare the value of a build to the normal value / range. In the comparison, the degree of variation relative to the normal value / range can be evaluated (e.g., how far a certain build is from the normal melting time, etc.).

[0151] For example, a build may be represented by a set of features (e.g., a set of one hundred builds may each be represented by ten features, etc.). The learner circuit 620 may construct a distribution for each feature and extract limits from each distribution. The learner circuit 620 learns that values ​​outside the limits are undesirable for the feature. The learner circuit 620 may also determine how far the value exceeds the limit.

[0152] The univariate model outputs can be fitted to the multivariate model outputs to determine which builds are standard, which builds are non-standard, and the severity or extent of the non-standard behavior. Features that contribute to non-standard behavior can be identified. Thus, behavior can be identified at the build level (e.g., which builds) as well as at the feature level (e.g., which features, distributions, rankings, etc.). Specific features can be isolated based on their impact on standard vs. non-standard build behavior. For example, hierarchical analysis can also be performed to form a median profile of key features.

[0153] In some examples, unsupervised learning is used to train multivariate models (e.g., no prior knowledge about what is good, what is bad, etc.). Techniques such as K-NN anomaly detection, HDBScan clustering, etc. can be used to identify and cluster signs or features of certain behaviors. Anomaly scores can be calculated (e.g., at least two different scores, etc.). For example, ten nearest neighbors can be identified, and the distance between neighbors is related to a score (1, 10, etc.) based on the distance to the nearest neighbor with anomalies. Builds with high anomaly scores are clustered or classified as non-standard builds. In order to classify or separate such builds from other builds, HDBScan clustering can then be used to filter out noise and cluster reasonably similar builds. If not similar, other builds will be considered noise by HDBScan. Clusters of reasonably similar builds can be formed, and the parameters or characteristics that define the clusters can become thresholds for evaluating builds and determining whether the build fits a certain cluster (e.g., non-standard or standard). These thresholds are saved to evaluate new builds later.

[0154] If a build is standard, its associated anomaly score is lower (e.g., 1, 2, etc.). Thus, if the anomaly score is less than 2, the build is classified or considered a standard build. Standard build regions and non-standard build regions are defined by looking at all features simultaneously. For example, if a build does not fall into either region (standard or non-standard), the build can be flagged for further review.

[0155] At block 1140, the learner circuit 620 classifies the build based on the learning and measured behaviors of the historical data set. The classification may form a set of standard build behaviors and a set of non-standard build behaviors. A severity or degree or measure of behavior contribution may be associated with a build, a build feature, etc. For example, one feature may contribute more to a standard or non-standard behavior of a build than another feature.

[0156] At block 1150, reference standards and non-standard behaviors are generated and saved for later use. For example, based on the classification and scoring of the features of the build population, the evaluator circuit 630 can generate and store "gold standard", normal or reference behaviors for training composite models and applications. Similarly, the learner circuit 620 can generate and store non-standard, abnormal or anomalous reference behaviors based on the classification and scoring of the features of the build population. Like the standard reference behaviors, the non-standard reference behaviors can be used to train the composite model and applied by the evaluator circuit 630 to the received build data.

[0157] At box 1160, a composite model is generated from standard and non-standard reference behaviors. For example, the composite model can be an AI model that is trained and tested using standard and non-standard reference behaviors, so that the composite model can distinguish between standard and non-standard behaviors, and, for example, identify non-standard behaviors in new builds. By training and testing the composite model using known / validated standard and non-standard behaviors, the composite model can be used to classify new builds as standard or non-standard based on the quantified behaviors reflected in the composite model. In some examples, machine parameters that cause such behavior can also be identified by the composite model based on its training from the identified behaviors and associated build data. For example, identifying machine parameters and / or parameter values ​​may be important for correcting non-standard build / machine behaviors.

[0158] Thus, the trained, tested, and deployed composite models can identify and explain variations at the build level, system level, subsystem level (e.g., bundle subsystem, etc.), etc. The composite models can be deployed on various populations (e.g., current build, builds in the past six months, builds in the past year, etc.). The composite models can be trained and tested for various AM machine populations (e.g., DMLM, EBM, BinderJet, etc.).

[0159] At block 1170, the composite model is deployed to the evaluator circuit 630 and / or otherwise stored, output, etc. For example, the composite model and / or associated standard and non-standard behavior reference information may be stored in the data storage device 640, 740 for use by the evaluator circuit 630. The evaluator circuit 630 may use the composite model to classify a build as standard or non-standard, identify contributing / causal factors for a non-standard build, and output corrective actions for the non-standard build.

[0160] Fig.12 Further example details regarding the evaluation of a build by the example evaluator circuit 630 described above are provided. At block 1210, build data is ingested by the evaluator circuit 630. For example, build configuration / setup information, sensor data, images, control events, build metadata, images before and after recoating, melt pool time series data, log files, other sensor values, etc. are provided to the evaluator circuit 630 directly and / or through the data storage device 610. The evaluator circuit 630 can process the raw sensor values, images, machine settings, build parameters, etc. to form features (e.g., using the feature engine 612) for processing by the composite model.

[0161] For example, feature extractor 612 may extract one or more features from settings of evaluator circuit 630, sensor statistics, build events, physical / domain based features, etc. Functionality may be related to events, filtering, inerting, laser firing, repainting, etc. Features may be based on values, attributes, etc. Features may be based on maximum / minimum values, average values, time weighted average values, specific record values, standard deviations, etc.

[0162] At block 1220, the evaluator circuit 630 evaluates the build against the reference behavior. For example, the build data is evaluated at the build level, layer level, and / or component level relative to the reference behavior stored in the data storage device 640, 740. One or more features may be evaluated and compared to multivariate and univariate aspects of the composite model. For example, a score (e.g., anomaly score, etc.) may be calculated that represents the distance between the features of the reviewed build and the corresponding features of the composite model.

[0163] At block 1230, the evaluator circuit 630 generates one or more scores or metrics for the build. For example, the evaluator circuit 630 generates a non-compliance severity metric. Multiple anomaly scores can be combined into a severity metric, indicating a non-compliance build and / or a non-compliance with respect to a particular feature, setting, characteristic, etc. In some examples, attribute-based severity scores are summed (e.g., the more the associated region / range is exceeded, the higher the severity of the defect) to form a build-level severity score. For example, a set of five features can be combined for two builds to determine severity.

[0164] At block 1240, the build is classified as standard or non-standard based at least in part on the score / metric. For example, when the build has a low non-compliance severity metric (e.g., 0.05, 0.1, etc.), then the build is classified as a standard build. When the build has a high non-compliance severity metric, then the build is classified as a non-standard build.

[0165] For example, a low non-compliance severity metric of 3.0 may be enforced to label a build as "absolutely standard". Conversely, a high non-compliance severity metric of 10.0 may be used as a guideline for identifying anomalous builds. In some examples, other metrics are further relied upon to obtain a "non-standard build" label. For example, a build may be evaluated from a reference frame containing 25 features, for a total non-compliance severity metric of 4.5, where 3 features (e.g., melt duration, rake duration, column temperature) contribute 2.1, 1.3, and 1.2 points to the total non-compliance severity metric, respectively. For example, such a score may be considered standard or undetermined.

[0166] The range or threshold of the non-compliance severity metric often depends on the "non-standard behavior" being modeled. For example, multiple features in a previously unseen combination (e.g., three features, four features, etc.) can drive the determination of non-standard behavior, as reflected in a high non-compliance severity metric. For example, an outlier value for a physical feature may drive a high non-compliance severity metric. This intent influences the choice of parameters / features, which in turn leads to the selection of the best threshold based on the specific situation. The threshold for non-compliance severity can be selected in combination with other classification methods (e.g., KNN-based anomaly scores and regions of noise-free clustered data from HDBScan methods, etc.).

[0167] At block 1250, when a build is classified as a non-standard build, one or more features / factors that contribute to the determination of a non-standard build are identified. For example, one or more features having a non-compliant severity metric and / or anomaly score above a threshold may be identified as contributing factors to the classification of a non-standard build.

[0168] At block 1260, a classification of the build is reported. For example, an output is generated indicating that the build has been classified as a standard or non-standard build. For example, the output may include an indication of the severity of the non-standard build. The indication of severity may also be associated with one or more features that cause and / or otherwise contribute to the build being classified as non-standard. Thus, the output classification of the build may include an overall assessment of the build, one or more build and / or feature severity scores, a hierarchy graph, and the like. In some examples, a non-compliance severity metric may be calculated, such as at a feature level, to quantify changes in the behavior of the build (e.g., changes in the behavior of the associated additive machine 100, 602 during the build, etc.).

[0169] At box 1270, if the build is classified as a non-standard build, a corrective action is output. For example, adjustments to one or more parameters of the ongoing build can be provided to the associated additive machine 100, 602, the associated controller 120, etc. In some examples, predictive maintenance can be scheduled to correct problems identified in the additive machine 100, 602 that cause non-standard builds (or, for example, builds that are barely standard but tend toward non-standard build behavior). In some examples, depending on the severity of the abnormal / non-standard build, a command can be sent to cancel the build. In some examples, an interactive display is provided in which the user can visibly observe the build score and associated features so that the user can observe, select, and understand which features contribute to non-standard behavior. For example, adjustments can then be made via a graphical user interface.

[0170] In some examples, at block 1280 , feedback may be collected based on continuation of a build after parameter / setting adjustments, post-mortem review of a canceled build, user overrides, etc. For example, such feedback may modify the behavior of the evaluator circuit 630 .

[0171] Fig.13 is structured to execute Figure 10-12 620, evaluator circuit 630, etc. The processor platform 1300 may be, for example, a server, a personal computer, a workstation, a self-learning machine (e.g., a neural network), a mobile device (e.g., a cell phone, a smartphone, a tablet computer, such as an iPad TM ), personal digital assistant (PDA), Internet appliance, or any other type of computing device.

[0172] The processor platform 1300 of the illustrated example includes a processor 1312. The processor 1312 of the illustrated example is hardware. For example, the processor 1312 may be implemented by one or more integrated circuits, logic circuits, microprocessors, GPUs, DSPs, or controllers from any desired family or manufacturer. The hardware processor may be a semiconductor-based (e.g., silicon-based) device.

[0173] The processor 1312 of the illustrated example includes a local memory 1313 (e.g., cache and / or other memory circuits). The processor 1312 of the illustrated example communicates with a main memory / memory circuit including a volatile memory 1314 and a non-volatile memory 1316 via a bus 1318. The volatile memory 1314 may be comprised of synchronous dynamic random access memory (SDRAM), dynamic random access memory (DRAM), Dynamic Random Access Memory The non-volatile memory 1316 may be implemented by flash memory and / or any other desired type of memory device / memory circuit. Access to the main memory 1314, 1316 is controlled by a memory controller.

[0174] The processor platform 1300 of the illustrated example also includes an interface circuit 1320. The interface circuit 1320 may be implemented by any type of interface standard (e.g., Ethernet interface, Universal Serial Bus (USB), interface, near field communication (NFC) interface and / or PCI express interface).

[0175] In the example shown, one or more input devices 1322 are connected to the interface circuit 1320. The input devices 1322 allow a user to enter data and / or commands into the processor 1312. The input devices may be implemented by, for example, an audio sensor, a microphone, a camera (still or video), a keyboard, buttons, a mouse, a touch screen, a trackpad, a trackball, and / or a voice recognition system.

[0176] One or more output devices 1324 are also connected to the interface circuit 1320 of the illustrated example. The output device 1324 can be implemented, for example, by a display device (e.g., a light emitting diode (LED), an organic light emitting diode (OLED), a liquid crystal display (LCD), a cathode ray tube display (CRT), an in-place switch (IPS) display, a touch screen, etc.), a tactile output device, and / or a speaker. Therefore, the interface circuit 1320 of the illustrated example typically includes a graphics driver card, a graphics driver chip, and / or a graphics driver processor.

[0177] The interface circuitry 1320 of the illustrated example also includes a communication device (e.g., a transmitter, a receiver, a transceiver, a modem, a residential gateway, a wireless access point, and / or a network interface) to facilitate data exchange with an external machine (e.g., any kind of computing device) via a network 1326. Communications may be via, for example, an Ethernet connection, a digital subscriber line (DSL) connection, a telephone line connection, a coaxial cable system, a satellite system, a field wireless system, a cellular system, etc.

[0178] The processor platform 1300 of the illustrated example also includes one or more mass storage devices 1328 for storing software and / or data. Examples of such mass storage devices 1328 include floppy disk drives, hard disk drives, compact disk drives, Blu-ray disk drives, redundant array of independent disks (RAID) systems, and digital versatile disk (DVD) drives.

[0179] Figure 10-12 The machine executable instructions 1332 may be stored in the mass storage device 1328, the volatile memory 1314, the non-volatile memory 1316, and / or on a removable non-transitory computer-readable storage medium (such as a CD or DVD).

[0180] Fig.14 yes Fig.13 1312 of the processor circuit 1312. In this example, Fig.13The processor circuit 1312 is implemented by the microprocessor 1400. For example, the microprocessor 1400 may implement a multi-core hardware circuit (such as a CPU, DSP, GPU, XPU, etc.). Although it may include any number of example cores 1402 (e.g., 1 core), the microprocessor 1400 of this example is a multi-core semiconductor device including N cores. The cores 1402 of the microprocessor 1400 may operate independently or may cooperate to execute machine-readable instructions. For example, machine code corresponding to a firmware program, an embedded software program, or a software program may be executed by one of the cores 1402, or may be executed by multiple of the cores 1402 at the same time or at different times. In some examples, the machine code corresponding to the firmware program, the embedded software program, or the software program is divided into threads and executed in parallel by two or more of the cores 1402. The software program may correspond to Figure 10-12 A flowchart may represent a portion or all of machine-readable instructions and / or operations.

[0181] The core 1402 can communicate via an example bus 1404. In some examples, the bus 1404 can implement a communication bus to enable communications associated with one or more of the cores 1402. For example, the bus 1404 can implement at least one of an inter-integrated circuit (I2C) bus, a serial peripheral interface (SPI) bus, a PCI bus, or a PCIe bus. Additionally or alternatively, the bus 1404 can implement any other type of computing or electrical bus. The core 1402 can obtain data, instructions, and / or signals from one or more external devices via an example interface circuit 1406. The core 1402 can output data, instructions, and / or signals to one or more external devices via the interface circuit 1406. Although the core 1402 of this example includes an example local memory 1420 (e.g., a level 1 (L1) cache that can be divided into an L1 data cache and an L1 instruction cache), the microprocessor 1400 also includes an example shared memory 1410 (e.g., a level 2 (L2 cache)) that can be shared by the cores for high-speed access to data and / or instructions. Data and / or instructions may be transferred (eg, shared) by writing to and / or reading from the shared memory 1410. The local memory 1420 of each core 1402 and the shared memory 1410 may be a plurality of levels of cache memory and main memory (eg, Fig.13 1314, 1316). Typically, higher levels of memory in the hierarchy exhibit shorter access times and have smaller storage capacities than lower levels of memory. Changes to different levels of the cache hierarchy are managed (e.g., coordinated) by a cache coherence policy.

[0182] Each core 1402 may be referred to as a CPU, DSP, GPU, etc., or any other type of hardware circuit. Each core 1402 includes a control unit circuit 1414, an arithmetic and logic (AL) circuit (sometimes referred to as an ALU) 1416, a plurality of registers 1418, an L1 cache 1420, and an example bus 1422. Other structures may exist. For example, each core 1402 may include a vector unit circuit, a single instruction multiple data (SIMD) unit circuit, a load / store unit (LSU) circuit, a branch / jump unit circuit, a floating point unit (FPU) circuit, etc. The control unit circuit 1414 includes a semiconductor-based circuit that is configured to control (e.g., coordinate) data movement within the corresponding core 1402. The AL circuit 1416 includes a semiconductor-based circuit that is configured to perform one or more mathematical and / or logical operations on data within the corresponding core 1402. The AL circuit 1416 of some examples performs integer-based operations. In other examples, the AL circuit 1416 also performs floating-point operations. In yet other examples, AL circuit 1416 may include a first AL circuit that performs integer-based operations and a second AL circuit that performs floating-point operations. In some examples, AL circuit 1416 may be referred to as an arithmetic logic unit (ALU). Register 1418 is a semiconductor-based structure to store data and / or instructions, such as the results of one or more operations performed by AL circuit 1416 of corresponding core 1402. For example, register 1418 may include vector registers, SIMD registers, general registers, flag registers, segment registers, machine-specific registers, instruction pointer registers, control registers, debug registers, memory management registers, machine check registers, etc. Fig.14 As shown, registers 1418 may be arranged in blocks. Alternatively, registers 1418 may be organized in any other arrangement, format, or structure, including being distributed throughout core 1402 to reduce access time. Bus 1422 may implement at least one of an I2C bus, an SPI bus, a PCI bus, or a PCIe bus.

[0183] Each core 1402, and / or more generally, the microprocessor 1400 may include additional and / or alternative structures to those shown and described above. For example, there may be one or more clock circuits, one or more power supplies, one or more power gates, one or more cache home agents (CHA), one or more aggregation / common grid stops (CMS), one or more shifters (e.g., barrel shifters) and / or other circuits. The microprocessor 1400 is a semiconductor device that is made to include many interconnected transistors to implement the above structure in one or more integrated circuits (ICs) contained in one or more packages. The processor circuit may include one or more accelerators and / or collaborate with one or more accelerators. In some examples, the accelerator is implemented by a logic circuit to perform certain tasks faster and / or more efficiently than a general-purpose processor can accomplish. Examples of accelerators include ASICs and FPGAs, such as those discussed herein. A GPU or other programmable device may also be an accelerator. The accelerator may be on a processor circuit, in the same chip package as the processor circuit, and / or in one or more packages separate from the processor circuit.

[0184] Fig.15 yes Fig.13 1312 is a block diagram of another example implementation of a processor circuit 1312. In this example, the processor circuit 1312 is implemented by an FPGA circuit 1500. The FPGA circuit 1500 can, for example, be used to execute a program that can otherwise be executed by executing corresponding machine-readable instructions. Fig.14 However, once constructed, FPGA circuit 1500 instantiates machine-readable instructions in hardware and can therefore typically perform operations faster than those performed by a general-purpose microprocessor executing corresponding software.

[0185] More specifically, Fig.14 The microprocessor 1400 (which can be programmed to execute Figure 10-12 In contrast, a general purpose device whose interconnections and logic circuits are fixed once fabricated is a flowchart that represents some or all of the machine-readable instructions. Fig.15 The example FPGA circuit 1500 includes interconnects and logic circuits, which may be constructed and / or interconnected in different ways after fabrication to instantiate, for example, Figure 10-12In particular, FPGA 1500 can be thought of as an array of logic gates, interconnects, and switches. The switches can be programmed to change how the logic gates are interconnected by the interconnects, effectively forming one or more dedicated logic circuits (unless and until FPGA circuit 1500 is reprogrammed). The constructed logic circuits enable the logic gates to cooperate in different ways to perform different operations on the data received by the input circuits. These operations can correspond to the Figure 10-12 Thus, FPGA circuit 1500 can be constructed to effectively Figure 10-12 Some or all of the machine-readable instructions of the flowchart of the instantiation of the dedicated logic circuit, so as to perform the operations corresponding to those software instructions in a dedicated manner similar to an ASIC. Figure 10-12 The operation of some or all of the machine-readable instructions of a processor faster than a general-purpose microprocessor can perform the same operations.

[0186] exist Fig.15 In the example of FPGA circuit 1500, FPGA circuit 1500 is constructed to be programmed (and / or reprogrammed one or more times) by an end user via a hardware description language (HDL) such as Verilog. Fig.15 FPGA circuit 1500 includes example input / output (I / O) circuit 1502 to obtain data from and / or output data to example construction circuit 1504 and / or external hardware (e.g., external hardware circuit) 1506. For example, construction circuit 1504 can implement interface circuits that can obtain machine-readable instructions to construct FPGA circuit 1500 or a portion thereof. In some such examples, construction circuit 1504 can obtain machine-readable instructions from a user, a machine (e.g., a hardware circuit (e.g., a programmed or dedicated circuit) that can implement an artificial intelligence / machine learning (AI / ML) model to generate instructions), etc. In some examples, external hardware 1506 can implement Fig.14 The FPGA circuit 1500 also includes an array of example logic gate circuits 1508, a plurality of example configurable interconnects 1510, and example storage circuits 1512. The logic gate circuits 1508 and the interconnects 1510 may be configured to instantiate a logic gate circuit that may correspond to Figure 10-12 One or more operations of at least some machine readable instructions and / or other desired operations. Fig.15The logic gate circuit 1508 shown in is made in the form of groups or blocks. Each block includes a semiconductor-based electrical structure that can be constructed into a logic circuit. In some examples, the electrical structure includes a logic gate (e.g., an AND gate, an OR gate, a NOR gate, etc.) that provides a basic building block for the logic circuit. An electrically controlled switch (e.g., a transistor) is present in each of the logic gate circuits 1508, thereby realizing the construction of the electrical structure and / or logic gate to form a circuit for the desired operation. The logic gate circuit 1508 may include other electrical structures, such as a lookup table (LUT), a register (e.g., a flip-flop or a latch), a multiplexer, etc.

[0187] The interconnect 1510 of the illustrated example is a conductive path, trace, via, etc. that may include an electrically controlled switch (e.g., a transistor), the state of which may be changed by programming (e.g., using an HDL instruction language) to activate or deactivate one or more connections between one or more logic gate circuits 1508 to program a desired logic circuit.

[0188] The storage circuit 1512 of the illustrated example is configured to store the results of one or more operations performed by corresponding logic gates. The storage circuit 1512 may be implemented by registers, etc. In the illustrated example, the storage circuit 1512 is distributed among the logic gate circuit 1508 to facilitate access and increase execution speed.

[0189] Fig.15 The example FPGA circuit 1500 also includes an example dedicated operation circuit 1514. In this example, the dedicated operation circuit 1514 includes a dedicated circuit 1516, which can be called to implement common functions to avoid the need to program those functions on site. Examples of such dedicated circuits 1516 include memory (e.g., DRAM) controller circuits, PCIe controller circuits, clock circuits, transceiver circuits, memories, and multiplier-accumulator circuits. There may be other types of dedicated circuits. In some examples, the FPGA circuit 1500 may also include an example general-purpose programmable circuit 1518, such as an example CPU 1520 and / or an example DSP 1522. Other general-purpose programmable circuits 1518 may be present in addition or alternatively, such as GPUs, XPUs, etc. that can be programmed to perform other operations.

[0190] Thus, the example FPGA circuit 1500 may be used for (re)alignment and / or calibration of multi-laser alignment, stitching, other aspects of additional build execution, programming, etc. In certain examples, the FPGA circuit 1500 may be used for scoring and data processing, along with and / or further in conjunction with hyper-recording of data / events where model outputs identify deviations from expected norms. For example, such additional data collected during hyper-recording may include additional sensor data and high frequency sub-second control outputs for troubleshooting and investigation.

[0191] although Fig.14 and 15 Shows Fig.13 These are two example implementations of the processor circuit 1312, but many other approaches are contemplated. For example, as described above, modern FPGA circuits may include an onboard CPU, such as Fig.15 One or more of the example CPUs 1520. Thus, Fig.13 The processor circuit 1312 may additionally be combined by Fig.14 An example microprocessor 1400 and Fig.15 In some such hybrid examples, the Figure 10-12 The first portion of the machine-readable instructions represented by the flowchart may be represented by Fig.14 1402 and is executed by one or more cores 1402 of Figure 10-12 The second portion of the machine-readable instructions represented by the flowchart may be represented by Fig.15 FPGA circuit 1500 executes.

[0192] Fig.16 A block diagram illustrating an example software distribution platform 1605 is shown in Fig.13 The example software distribution platform 1605 may be implemented by any computer server, data facility, cloud service, etc. that is capable of storing software and transferring software to other computing devices. The third party may be a customer of the entity that owns and / or operates the software distribution platform 1605. For example, the entity that owns and / or operates the software distribution platform 1605 may be a software (such as a hardware device owned and / or operated by a third party of the owner and / or operator of the software distribution platform). Fig.13 The third party may be a consumer, user, retailer, OEM, etc. who purchases and / or licenses the software for use and / or resale and / or sublicense. In the example shown, the software distribution platform 2105 includes one or more servers and one or more storage devices. The storage device stores the machine-readable instructions 1332, which may correspond to Figure 10-12The example machine-readable instructions of the example software distribution platform 1605 are described above. One or more servers of the example software distribution platform 1605 communicate with the example network 1610, which can correspond to any one or more of the Internet and / or any of the above-described example networks. In some examples, one or more servers respond to a request to transfer software to a requesting party as part of a business transaction. The delivery, sale, and / or payment of a license for the software can be processed by one or more servers of the software distribution platform and / or a third-party payment entity. The server enables a purchaser and / or a licensee to download the machine-readable instructions 1332 from the software distribution platform 1605. For example, the machine-readable instructions 1332 can be downloaded from the software distribution platform 1605. Figure 10-12 The example machine-readable instructions of the software are downloaded to the example programmable circuit platform 1300, which will execute the machine-readable instructions 1332 to implement all or part of the example additive manufacturing machine behavior device 600, etc. In some examples, one or more servers of the software distribution platform 1605 periodically provide, transmit and / or force updates of the software (e.g., Fig.13 Example machine readable instructions 1332 of the system to ensure that improvements, patches, updates, etc. are distributed and applied to the software on the end-user device. Although referred to as software above, the distributed "software" may alternatively be firmware.

[0193] It should now be understood that the devices, systems, and methods described herein monitor, determine, and adjust the health of additive manufacturing devices and / or related processes, builds, and the like. These systems and methods ingest data, manage status, process analysis, and generate actionable outputs for layering, build-side, and machine-side adjustments. Such monitoring, processing, and adjustments are not possible manually and require reliance on an analytical processor. In addition, the systems and methods according to the present disclosure improve the accuracy of diagnosing builds and / or additive manufacturing devices by determining and reacting to the specific and overall health of the machine, process, and / or build.

[0194] Further aspects are provided by the subject matter of the following clauses:

[0195] An example device includes a learner circuit, the learner circuit being used to: process first data from a set of first builds to learn behaviors from the set of first builds; classify each build in the set of first builds as a standard build or a non-standard build; model the learned behaviors to form a standard reference behavior and a non-standard reference behavior, the standard reference behavior including a first feature and the non-standard reference behavior including a second feature; and output the standard reference behavior and the non-standard reference behavior to classify additional builds. The example device includes an evaluator circuit, the evaluator circuit being used to: ingest second data of a second build; compare the second data with the standard reference behavior and the non-standard reference behavior; classify the second build as a standard build or a non-standard build; and when the second build is classified as a non-standard build, output a corrective action to address at least one second feature of the non-standard build behavior associated with the second build.

[0196] The apparatus of the preceding clause, further comprising a memory circuit for storing the standard reference behavior and the non-standard reference behavior.

[0197] An apparatus as described in any preceding clause, wherein the standard reference behavior and the non-standard reference behavior form a composite model, the composite model being deployed for use by the evaluator circuit to classify the second construction.

[0198] An apparatus as described in any preceding clause, wherein the learner circuit is used to construct and process at least one univariate model and at least one multivariate model to form the standard reference behavior and the non-standard reference behavior.

[0199] Apparatus as claimed in any preceding clause, wherein the set of first builds is from one or more additive manufacturing machines.

[0200] Apparatus as claimed in any preceding clause, wherein the second build is an ongoing build on an additive manufacturing machine.

[0201] Apparatus as in any preceding clause, wherein the first feature and the second feature comprise build-level features and layer-level features.

[0202] An apparatus as described in any preceding clause, wherein the learner circuit is used to calculate a non-compliance severity metric to classify a build as a standard build or a non-standard build, the non-compliance severity metric being based on scores associated with the first feature and the second feature, the non-compliance severity metric being capable of identifying one or more of the second features that contribute to the classification as a non-standard build.

[0203] An example non-transitory computer-readable medium comprising instructions that, when executed by a processor circuit, cause the processor circuit to at least: process first data from a set of first builds to learn behavior from the set of first builds; classify each build in the set of first builds as a standard build or a non-standard build; model the learned behavior to form a standard reference behavior and a non-standard reference behavior, the standard reference behavior including a first feature and the non-standard reference behavior including a second feature; output the standard reference behavior and the non-standard reference behavior to classify an additional build; ingest second data of a second build; perform comparison processing on the second data with the standard reference behavior and the non-standard reference behavior; classify the second build as a standard build or a non-standard build; and when the second build is classified as a non-standard build, output a corrective action to address at least one second feature of the non-standard reference behavior associated with the second build.

[0204] The non-transitory computer-readable medium of any preceding clause, wherein the processor circuit comprises a learner circuit and an evaluator circuit, the learner circuit being configured to store the standard reference behavior and the non-standard reference behavior for use by the evaluator circuit in classifying the second construction.

[0205] The non-transitory computer-readable medium of any preceding clause, wherein the standard reference behavior and the non-standard reference behavior form a composite model, the composite model being deployed to classify the second construct.

[0206] The non-transitory computer-readable medium of any preceding clause, wherein the processor circuit is configured to construct and process at least one univariate model and at least one multivariate model to form the standard reference behavior and the non-standard reference behavior.

[0207] The non-transitory computer-readable medium of any preceding clause, wherein the set of first builds is from one or more additive manufacturing machines.

[0208] The non-transitory computer-readable medium of any preceding clause, wherein the second build is an ongoing build on an additive manufacturing machine.

[0209] The non-transitory computer-readable medium of any preceding clause, wherein the first feature and the second feature comprise a build-level feature and a layer-level feature.

[0210] A non-transitory computer-readable medium as described in any preceding clause, wherein the processor circuit is used to calculate a non-compliance severity metric to classify a build as a standard build or a non-standard build, the non-compliance severity metric being based on scores associated with the first feature and the second feature, the non-compliance severity metric being capable of identifying one or more of the second features that contribute to the classification as a non-standard build.

[0211] An example method for analyzing and managing builds in one or more additive manufacturing machines includes: processing first data from a set of first builds by using a processor circuit to execute instructions to learn behavior from the set of first builds; classifying each build in the set of first builds as a standard build or a non-standard build by using the processor circuit to execute instructions; modeling the learned behavior to form a standard reference behavior and a non-standard reference behavior by using the processor circuit to execute instructions, the standard reference behavior including a first feature and the non-standard reference behavior including a second feature; outputting the standard reference behavior and the non-standard reference behavior to classify additive builds by using the processor circuit to execute instructions; ingesting second data of a second build by using the processor circuit to execute instructions; comparing the second data with the standard reference behavior and the non-standard reference behavior by using the processor circuit to execute instructions; classifying the second build as a standard build or a non-standard build by using the processor circuit to execute instructions; and when the second build is classified as a non-standard build, outputting a corrective action by using the processor circuit to execute instructions to address at least one second feature of the non-standard reference behavior associated with the second build.

[0212] A method as in any preceding clause, further comprising forming a composite model having the standard reference behavior and the non-standard reference behavior, the composite model being deployed to classify the second construct.

[0213] A method as described in any preceding clause, wherein classification further comprises constructing and processing at least one univariate model and at least one multivariate model to form the standard reference behavior and the non-standard reference behavior.

[0214] The method of any preceding clause, further comprising calculating a non-compliance severity metric to classify a build as a standard build or a non-standard build, the non-compliance severity metric being based on scores associated with the first feature and the second feature, the non-compliance severity metric being capable of identifying one or more of the second features that contribute to the classification as a non-standard build.

[0215] Although specific examples have been shown and described herein, it should be understood that various other changes and modifications may be made without departing from the spirit and scope of the claimed subject matter. In addition, although various aspects of the claimed subject matter are described herein, these aspects do not need to be used in combination. Therefore, the appended claims are intended to cover all such changes and modifications within the scope of the claimed subject matter.

Claims

1. A device, characterized in that: include: A learner circuit, the learner circuit being configured to: processing first data from a set of first builds to learn behavior from the set of first builds; classifying each build in the set of first builds as a standard build or a non-standard build; modeling the learned behavior to form a standard reference behavior and a non-standard reference behavior, the standard reference behavior including a first feature, and the non-standard reference behavior including a second feature; and outputting the standard reference behavior and the non-standard reference behavior to classify an additional build; and an evaluator circuit, the evaluator circuit being configured to: ingesting second data of the second build; Comparing the second data with the standard reference behavior and the non-standard reference behavior; classifying the second build as a standard build or a non-standard build; and When the second build is classified as a non-standard build, a corrective action is output to address at least one second characteristic of the non-standard build behavior associated with the second build.

2. The device according to claim 1, characterized in that Further included is a memory circuit for storing the standard reference behavior and the non-standard reference behavior.

3. The device according to claim 1, characterized in that in, The standard reference behavior and the non-standard reference behavior form a composite model that is deployed for use by the evaluator circuit to classify the second construction.

4. The device according to claim 1, characterized in that in, The learner circuit is used to construct and process at least one univariate model and at least one multivariate model to form the standard reference behavior and the non-standard reference behavior.

5. The device according to claim 1, characterized in that in, The set of first builds is from one or more additive manufacturing machines.

6. The device according to claim 1, characterized in that in, The second build is an ongoing build on the additive manufacturing machine.

7. The device according to claim 1, characterized in that in, The first features and the second features include build-level features and layer-level features.

8. The device according to claim 1, characterized in that in, The learner circuit is used to calculate a non-compliance severity metric to classify a build as a standard build or a non-standard build, the non-compliance severity metric being based on scores associated with the first feature and the second feature, the non-compliance severity metric being capable of identifying one or more of the second features that contribute to the classification as a non-standard build.

9. A non-transitory computer-readable medium comprising instructions, characterized in that: When executed by a processor circuit, the instructions cause the processor circuit to at least: processing first data from a set of first builds to learn behavior from the set of first builds; classifying each build in the set of first builds as a standard build or a non-standard build; Modeling the learned behavior to form a standard reference behavior and a non-standard reference behavior, the standard reference behavior including a first feature, and the non-standard reference behavior including a second feature; outputting the standard reference behavior and the non-standard reference behavior to classify the additional construction; ingesting second data of the second build; Comparing the second data with the standard reference behavior and the non-standard reference behavior; classifying the second build as a standard build or a non-standard build; and When the second build is classified as a non-standard build, a corrective action is output to address at least one second characteristic of the non-standard reference behavior associated with the second build.

10. The non-transitory computer readable medium of claim 9, wherein: in, The processor circuit includes a learner circuit and an evaluator circuit, the learner circuit being configured to store the standard reference behavior and the non-standard reference behavior for use by the evaluator circuit in classifying the second construction.

Images