Laser powder bed fusion additive manufacturing anomaly detection

Abstract

This disclosure provides techniques for detecting anomalies in additive manufacturing. An example method includes obtaining a time-series signal comprising multiple samples from an electromagnetic energy sensor, wherein the samples correspond to electromagnetic emissions generated by the interaction between a first energy source and powder at a first focal point on an additive material layer, and generating the samples at a sampling rate greater than or equal to a first sampling rate; decomposing the time-series signal into at least high-frequency signal variation components; calculating one or more spatiotemporal statistics based on the high-frequency signal variation components of the time-series signal; identifying one or more disturbances from the one or more spatiotemporal statistics using a machine learning model, the one or more disturbances exceeding predetermined thresholds corresponding to the one or more spatiotemporal statistics; and generating an alarm in response to the identification of the one or more disturbances.

Description

Technical Field

[0001] This specification relates to systems and methods for detecting anomalies in additive manufacturing. Background Technology

[0002] Compared to subtractive manufacturing methods, additive manufacturing (AM) processes typically involve the stacking of one or more materials to create net-shape or near-net-shape (NNS) objects. Although "additive manufacturing" is the industry-standard term (ISO / ASTM 52900), AM encompasses a wide range of manufacturing and prototyping techniques under various names, including freeform manufacturing, 3D printing, rapid prototyping / machining, and more. AM technologies enable the fabrication of complex parts using a variety of materials. Often, freestanding objects can be manufactured from computer-aided design (CAD) models.

[0003] A particular type of AM process uses an energy source, such as an irradiation-guided device, to direct an energy beam, such as an electron beam or laser beam, to sinter or melt powder materials, creating solid three-dimensional objects by binding the powder material particles together. AM processes may use different material systems or additive powders, such as engineering plastics, thermoplastic elastomers, metals, and / or ceramics. Laser sintering or melting is a prominent AM process used for the rapid fabrication of functional prototypes and tooling. Applications include the direct fabrication of complex workpieces, patterning in investment casting, metal molds for injection molding and die casting, and molds and cores for sand casting. Fabricating prototype objects to enhance communication and testing of concepts during the design cycle is another common use of AM processes.

[0004] To monitor the additive manufacturing process, some conventional additive manufacturing machines include a melt pool monitoring system. Quality assessment can be used to adjust the build process, stop the build process, eliminate build process anomalies, issue warnings to machine operators, and / or identify suspicious or poor-quality parts produced during the build process.

[0005] Additive manufacturing machines with improved anomaly detection systems would be useful. More specifically, systems and methods utilizing molten pool monitoring systems configured for anomaly detection would be particularly beneficial. Attached Figure Description

[0006] The aspects illustrated in the accompanying drawings are illustrative and exemplary in nature and are not intended to limit this disclosure. The following detailed description of the illustrative aspects can be understood when read in conjunction with the following drawings, wherein like structures are indicated by like reference numerals.

[0007] Figure 1 An illustrative schematic diagram of an additive manufacturing machine according to one or more aspects shown and described herein is depicted.

[0008] Figure 2 Depicting one or more aspects as shown and described herein Figure 1 An illustrative close-up diagram of a construction platform for an exemplary additive manufacturing machine.

[0009] Figure 3 A flowchart is provided illustrating an illustrative method for detecting anomalies in additive manufacturing, based on one or more aspects shown and described herein.

[0010] Figure 4 Exemplary high-rate time-series signals are depicted according to one or more aspects shown and described herein.

[0011] Figure 5 An illustrative system for implementing anomaly detection in additive manufacturing, according to one or more aspects shown and described herein, is described. Detailed Implementation

[0012] This disclosure provides systems and methods for detecting anomalies in additive manufacturing. More specifically, these aspects are intended to provide systems and methods for detecting anomalies (also known as defects) using high-frequency analysis of molten pool emission signals.

[0013] As described in detail herein, aspects of this disclosure provide a technical solution to the technical problems associated with conventional methods of molten pool monitoring systems, which miss many anomalies of interest because the anomalies are too small to affect the intensity or temperature of the interaction between the energy source and the powder at the first focal point on the additive layer. Conventional methods for implementing molten pool monitoring systems may also miss anomalies of interest because the interaction time corresponding to the anomaly may be too short to affect the intensity or temperature.

[0014] To address these issues, the additive manufacturing anomaly detection technology described in this paper combines the collection of high-speed time-series signals with the analysis of high-frequency signal variation components to overcome the technical problem of missed anomaly detection in traditional molten pool monitoring systems. As described in more detail herein, the additive manufacturing anomaly detection technology can detect small anomalies (e.g., anomalies of 1 mm or smaller) and anomalies occurring on short timescales (e.g., on microsecond or smaller scales) based on high-speed time-series signals collected by electromagnetic energy sensors. The high-speed time-series signals are decomposed into high-frequency signal variation components. One or more spatiotemporal statistics can be calculated based on the high-frequency signal variation components and analyzed using a machine learning model configured to identify one or more disturbances from the one or more spatiotemporal statistics that exceed predetermined thresholds corresponding to the one or more spatiotemporal statistics.

[0015] As used herein, the term "spatiotemporal" refers to the spatial and temporal relationship between a signal and its corresponding data value. That is, as discussed in more detail herein, additive manufacturing processes can build objects layer by layer. A toolpath defined for each layer being built defines the pattern followed by the material deposition tool and / or energy source as it builds the object. For example, the toolpath can include a back-and-forth pattern (also known as a raster toolpath), an offset pattern (e.g., a tool trajectory that starts from the outside or inside of the object and moves inward or outward, respectively), or other similar techniques. In aspects of this disclosure, an electromagnetic energy sensor generates a high-rate time-series signal corresponding to the toolpath, thereby enabling the acquisition of a high-rate time-series signal corresponding to the electromagnetic emission generated by the interaction between the energy source and the powder at a first focal point on the additive material layer. Therefore, as the generated time-series signal, each data value is spatially and temporally correlated with other data values.

[0016] Additive manufacturing anomaly detection technology relies on electromagnetic energy sensors in additive manufacturing monitoring systems. These sensors are capable of collecting data at high rates, such as 10 kHz or higher, 20 kHz or higher, 30 kHz or higher, 40 kHz or higher, 50 kHz or higher, 60 kHz or higher, 70 kHz or higher, 80 kHz or higher, 90 kHz or higher, 100 kHz or higher, or 200 kHz or higher. In some aspects, the sampled data can be downsampled to a lower sampling rate after collection. Electromagnetic energy sensors enable the acquisition of high-rate time-series signals, which correspond to electromagnetic emissions generated by the interaction between the energy source and the powder at a first focal point on the additive material layer.

[0017] Additive manufacturing anomaly detection technology can be applied to various additive manufacturing processes. Additive manufacturing encompasses various manufacturing and prototyping techniques, including freeform manufacturing, 3D printing, rapid prototyping / machining, etc. Additive manufacturing technology can manufacture complex parts using a variety of materials.

[0018] Selective laser sintering, direct laser sintering, selective laser melting, and direct laser melting are common industry terms used to refer to the production of three-dimensional (3D) objects by using a laser beam to sinter or melt fine powders. More precisely, sintering requires melting (coagulating) powder particles at temperatures below the melting point of the powder material, while melting requires completely melting the powder particles to form a solid homogeneous substance. The physical processes associated with laser sintering or laser melting involve transferring heat to the powder material and then sintering or melting it. Although laser sintering and melting processes can be applied to a wide range of powder materials, the scientific and technical aspects of the production route, such as the sintering or melting rate and the influence of processing parameters on the microstructure evolution during layer fabrication, are not well understood. This manufacturing method involves multiple forms of heat, mass, and momentum transfer, as well as chemical reactions that make the process highly complex.

[0019] During direct metal laser sintering (DMLS) or direct metal laser melting (DMLM), the apparatus constructs an object layer by layer by sintering or melting powder material using an energy beam. The powder to be melted by the energy beam is uniformly distributed on a powder bed on the build platform, and the energy beam, under the control of an irradiation beam guide, sintersects or melts cross-sectional layers of the object being constructed. The build platform is lowered, another powder layer is laid on the powder bed and the object being constructed, and then the powder is continuously melted / sintered. This process is repeated until the part is entirely composed of molten / sintered powder material.

[0020] After the parts are manufactured, various post-processing procedures can be applied. Post-processing procedures include removing excess powder by means such as purging or vacuuming. Other post-processing procedures include stress relief processes. In addition, thermal, mechanical, and chemical post-processing procedures can be used to finish the parts.

[0021] To monitor the additive manufacturing process, additive manufacturing machines include additive manufacturing monitoring systems. These systems typically include one or more electromagnetic energy sensors, such as cameras, photodiodes, pyrometers, and light sensors, to detect electromagnetic emissions generated by the molten pool from an energy beam or otherwise emitted (e.g., electromagnetic emissions resulting from the interaction between the energy source and the powder at a first focal point on the additive material layer). Camera or sensor values ​​can be used to assess the quality of the build during or after the build process. Quality assessments can be used to adjust the build process, stop the build process, eliminate build process anomalies, warn the machine operator, and / or identify questionable or poor-quality parts produced by the build.

[0022] Reference will now be made in detail to aspects of the invention, with one or more examples of the invention illustrated in the accompanying drawings. Each example is provided by way of explanation and not limitation of the invention. Indeed, those skilled in the art will understand that various modifications and variations can be made to this disclosure without departing from the scope or spirit of the invention. For example, a feature shown or described as part of one aspect may be used with another aspect to produce yet another aspect. Therefore, this disclosure is intended to cover modifications and variations within the scope of the appended claims and their equivalents.

[0023] As used herein, the terms “first,” “second,” and “third” are used interchangeably to distinguish one component from another and do not imply the position or importance of the components. Furthermore, as used herein, approximate terms such as “approximate,” “substantially,” or “about” refer to an error within 10%.

[0024] Systems and methods for detecting anomalies in additive manufacturing utilize an additive manufacturing monitoring system coupled to an additive manufacturing machine to acquire a high-rate time-series signal. This high-rate time-series signal corresponds to electromagnetic emissions generated by the interaction between an energy source and powder at a first focal point on the additive material layer. The high-rate time-series signal is decomposed into at least high-frequency signal variation components. One or more spatiotemporal statistics can be calculated based on these high-frequency signal variation components. A machine learning model analyzes these one or more spatiotemporal statistics to identify one or more perturbations exceeding predetermined thresholds corresponding to these statistics.

[0025] See Figure 1 A laser powder bed melting system, referred to herein as "AM system 100," such as a DMLS or DMLM system, will be described according to exemplary aspects. As shown, AM system 100 may include a fixed housing 102 that provides a contaminant-free and controlled environment for performing additive manufacturing processes. For example, housing 102 is used to isolate and protect other components of AM system 100. Furthermore, housing 102 may be equipped with a suitable protective gas flow, such as nitrogen, argon, or another suitable gas or gas mixture. Housing 102 may have a gas inlet port 104 and a gas outlet port 106 for receiving gas flows to generate a static pressurized volume or a dynamic gas flow.

[0026] The housing 102 may typically contain some or all of the components of the AM system 100. In some aspects, the AM system 100 typically includes a worktable 110 located within the housing 102, a powder supply device 112, a doctor blade or recoating mechanism 114, an overflow container or overflow reservoir 116, and a build platform 118. Furthermore, an energy source 120 generates an energy beam 122, and a beam manipulation device 124 guides the energy beam 122 to facilitate the AM process, as described in more detail below. Each of these components will be described in more detail below.

[0027] The workbench 110 is a rigid structure defining a planar build surface 130. Furthermore, the planar build surface 130 defines a build opening 132 through which a build chamber 134 can be accessed. In some respects, the build chamber 134 is at least partially defined by a vertical wall 136 and a build platform 118. Additionally, the build surface 130 defines a supply opening 140 and a reservoir opening 144 through which additive powder 142 can be supplied from a powder supply device 112, and excess additive powder 142 can enter an overflow reservoir 116 through the reservoir opening 144. The collected additive powder may optionally be processed to sieve out loose, agglomerated particles before reuse.

[0028] The powder supply device 112 may include an additive powder supply container 150, which typically contains a volume of additive powder 142 sufficient for some or all of the additive manufacturing processes for one or more specific parts. Furthermore, the powder supply device 112 includes a supply platform 152, which is a plate-like structure movable vertically within the additive powder supply container 150. More specifically, a supply actuator 154 vertically supports the supply platform 152 and selectively moves it up and down during the additive manufacturing process.

[0029] AM system 100 also includes a recoating mechanism 114, which is a rigid, laterally elongated structure adjacent to the build surface 130. For example, the recoating mechanism 114 may be a hard scraper, a soft scraper, or a roller. The recoating mechanism 114 is operatively coupled to a recoating actuator 160, which is operable to selectively move the recoating mechanism 114 along the build surface 130. Furthermore, a platform actuator 164 is operatively coupled to a build platform 118 and is generally operable to move the build platform 118 vertically during the build process. While actuators 154, 160, and 164 are shown as hydraulic actuators, it should be understood that, depending on the alternative aspects, any other type and configuration of actuators may be used, such as pneumatic actuators, hydraulic actuators, ball screw linear electric actuators, or any other suitable vertical support device. Other configurations are also possible and are within the scope of this subject matter.

[0030] Energy source 120 may include any known device operable to generate a beam of light with appropriate power and other operating characteristics to melt and molten metal powder during the construction process. For example, energy source 120 may be a laser. Other directional energy sources, such as electron beam guns, are suitable alternatives to lasers.

[0031] In some aspects, the beam manipulation device 124 includes one or more mirrors, prisms, lenses, and / or electromagnets operably coupled to suitable actuators and arranged to guide and focus the energy beam 122. For example, the beam manipulation device 124 may be a galvanometer scanner that moves or scans the focal point of the laser beam (e.g., energy beam 122) emitted by the energy source 120 on the build surface 130 during laser melting and sintering. Thus, the energy beam 122 can be focused to a desired spot size and manipulated to a desired position in a plane coinciding with the build surface 130. Galvanometer scanners in powder bed melting technology typically have a fixed position, but the movable mirrors / lenses included allow for control and adjustment of various characteristics of the laser beam. It should be understood that other types of energy sources 120 may be used, which may use and replace the beam manipulation device 124. For example, if the energy source 120 is an electronic control unit for guiding an electron beam, the beam manipulation device 124 may be, for example, a deflection coil.

[0032] Prior to the additive manufacturing process, the supply actuator 160 can be lowered to supply additive powder 142 of the desired composition (e.g., metal, ceramic, and / or organic powders) into the additive powder supply container 150. Furthermore, the platform actuator 164 can move the build platform 118 to an initial high position, for example, making it substantially flush with or coplanar with the build surface 130. The build platform 118 is then lowered below the build surface 130 in selected layer increments. The layer increment affects the speed of the additive manufacturing process and the resolution of the part 170 or component being manufactured. For example, the layer increment can be approximately 10 to 100 micrometers (0.0004 to 0.004 inches).

[0033] The additive powder is then deposited onto build platform 118 and melted by energy source 120. Specifically, supply actuator 154 can raise supply platform 152 to push powder through supply opening 140, exposing it above build surface 130. Recoating mechanism 114 can then move over build surface 130 via recoating actuator 160 to horizontally distribute the raised additive powder 142 onto build platform 118 (e.g., in selected layer increments or thicknesses). As recoating mechanism 114 moves from left to right (e.g.) Figure 1(As shown), any excess additive powder 142 will fall into the overflow reservoir 116 through the reservoir opening 144. The recoating mechanism 114 can then be moved back to the starting position. The leveled additive powder 142 can be referred to as the "build layer" 172 (see...). Figure 2 Its exposed upper surface can be referred to as build surface 130. When the build platform 118 is lowered into the build chamber 134 during the build process, the build chamber 134 and the build platform 118 together surround and support a large amount of additive powder 142 and any part being built. This large amount of powder is often referred to as a "powder bed", and this particular type of additive manufacturing process can be referred to as a "powder bed process".

[0034] In the additive manufacturing process, a directional energy source 120 is used to melt a two-dimensional cross-section or layer of the part 170 being constructed. More specifically, an energy beam 122 is emitted from the energy source 120, and a beam manipulation device 124 is used to manipulate the focus 174 of the energy beam 122 on the exposed powder surface in a suitable pattern. A small portion of the exposed additive powder layer 142 surrounding the focus 174 is referred to herein as a “weld pool” or “melt pool” or “heat-affected zone” 176 (in... Figure 2 The molten pool 176 (best visible in the middle) is heated by an energy beam 122 to a temperature that allows it to sinter or melt, flow, and solidify. For example, the width of the molten pool 176 is approximately 100 micrometers (0.004 inches). This step may be referred to as melting additive powder 142.

[0035] The build platform 118 moves vertically downwards in layer increments, applying another layer of additive powder 142 with a similar thickness. The directional energy source 120 fires an energy beam 122 again, and the beam manipulation device 124 manipulates the focus 174 of the energy beam 122 on the exposed powder surface in an appropriate pattern. The exposed layer of additive powder 142 is heated by the energy beam 122 to allow it to sinter or melt, flow, and solidify within the top layer and the previously cured layer below. This cycle of moving the build platform 118, applying additive powder 142, and then guiding the energy beam 122 to melt the additive powder 142 is repeated until the entire part 170 is completed.

[0036] As described above, when the energy source 120 and beam manipulation device 124 guide the energy beam 122 (e.g., a laser beam or electron beam) onto the powder bed or build surface 130, the additive powder 142 is heated and begins to melt into the molten pool 176, where it can be melted to form the final part 170. It is noteworthy that the heated material emits electromagnetic energy (also referred to herein as “electromagnetic emission”) in the form of visible and invisible light. A portion of the directed energy beam is reflected back to the galvanometer scanner or beam manipulation device 124, and a portion is generally scattered in all other directions within the housing 102. Monitoring the emitted and / or reflected electromagnetic energy can be used to improve process monitoring and control. Exemplary systems for monitoring additive manufacturing processes, including two exemplary types of monitoring sensors, are described below according to exemplary aspects. However, systems for monitoring additive manufacturing processes may include one or more monitoring sensors, such as electromagnetic energy sensors.

[0037] Still referencing Figure 1 The additive manufacturing monitoring system 200 (optionally also referred to as a molten pool monitoring system) can be used in conjunction with the AM system 100 to monitor the molten pool 176, and the general manufacturing process will be described according to exemplary aspects of this subject matter. The additive manufacturing monitoring system 200 includes one or more electromagnetic energy sensors, such as cameras, photodiodes, pyrometers, optical sensors, etc., for measuring the amount of visible or invisible electromagnetic energy emitted from or reflected by the molten pool 176. More specifically, according to the illustrated aspect, the additive manufacturing monitoring system 200 includes one or more on-axis sensors 202 (e.g., on-axis optical sensors and / or molten pool sensors) and optionally one fixed off-axis sensor 204 (e.g., off-axis optical sensors and / or molten pool sensors). Each of these sensors 202, 204 will be described below according to exemplary aspects. However, it should be understood that the additive manufacturing monitoring system 200 may include any other suitable type, number, and configuration of sensors for detecting electromagnetic energy and other characteristics of the molten pool 176 or the process in general.

[0038] As used herein, a “beamline” or “on-axis” molten pool sensor (e.g., one of one or more on-axis sensors 202) refers to a sensor typically positioned along the path of energy beam 122. These sensors can monitor emitted and / or reflected light returning along the beam path. Specifically, when energy beam 122 forms molten pool 176, a portion of the emitted and reflected electromagnetic energy from molten pool 176 returns to energy source 120 along the same path. On-axis sensor 202 may include a beamsplitter 206 positioned along the beamline, which may include a coating for redirecting a portion of the electromagnetic energy to beamline sensing element 208. In this respect, sensing element 208 may be, for example, a photodiode, pyrometer, optical camera, infrared (IR) camera, or spectral sensor configured to measure electromagnetic energy in any spectrum, such as infrared (IR), ultraviolet (UV), visible light, etc. On-axis sensor 202 can measure any suitable parameters of the filtered reflected beam, such as intensity, frequency, wavelength, etc.

[0039] Furthermore, as used herein, a "fixed" or "off-axis" molten pool sensor 204 refers to a sensor that typically has a fixed position relative to the molten pool 176 and is used to measure the electromagnetic energy generated by the energy beam 122 and the molten pool 176 within a specified field of view. Additionally, the fixed molten pool sensor 204 may include any suitable device, such as a photodiode or an infrared (IR) camera. The off-axis molten pool sensor 204 may operate similarly to an on-axis molten pool sensor, but is not located on the beamline and includes a sensing element 208 typically configured to monitor the scattered electromagnetic energy from the molten pool 176.

[0040] In some aspects of this subject matter, the additive manufacturing monitoring system 200 may also include one or more filters 210 for filtering electromagnetic energy before it reaches the sensing elements 208 of the respective sensors 202, 204. For example, one or more filters 210 may remove wavelengths of the energy beam 122, such that the sensors 202, 204 monitor only the reflected electromagnetic energy. Alternatively, one or more filters 21 may be configured to remove other unwanted wavelengths to improve monitoring of the molten pool 176 or the process in general.

[0041] The additive manufacturing monitoring system 200 also includes a controller 220 operatively coupled to the on-axis optical sensor 202 and / or the off-axis optical sensor 204 for receiving signals corresponding to detected electromagnetic energy. The controller 220 may be a dedicated controller for the additive manufacturing monitoring system 200 or a system controller for operating the AM system 100. The controller 220 may include one or more memory devices and one or more processors, such as general-purpose or dedicated processors, for executing programming instructions or microcontroller code related to the additive manufacturing process or process monitoring. The memory may represent random access memory, such as DRAM, or read-only memory, such as ROM or FLASH. In some respects, the processor executes programming instructions stored in the memory. The memory may be a component separate from the processor or may be included in an onboard component within the processor. Alternatively, the controller 220 may be constructed without a processor, for example, using a combination of discrete analog and / or digital logic circuits (such as switches, amplifiers, integrators, comparators, flip-flops, AND gates, etc.) for control functions, rather than relying on software.

[0042] As described above, conventional melt pool monitoring systems miss many anomalies of interest because the anomalies are too small to affect the intensity or temperature of the interaction between the energy source and the powder at the first focal point on the additive layer. Furthermore, conventional melt pool monitoring methods may also miss anomalies of interest because the interaction time corresponding to the anomaly may be too short to affect the intensity or temperature. Therefore, aspects of this disclosure address systems and methods for anomaly detection in additive manufacturing, which combine the collection of high-rate time-series signals from one or more electromagnetic energy sensors with the analysis of high-frequency signal variation components to solve the technical problem of anomaly detection missed by conventional melt pool monitoring systems.

[0043] Figure 3 A flowchart of a method 300 for anomaly detection in additive manufacturing is depicted. In some aspects, method 300 can be performed by an apparatus, such as referenced in [reference]. Figure 1 The controller 220 of the AM system 100 described and illustrated Figure 5 The computing system 500.

[0044] The method 300 for detecting anomalies in additive manufacturing begins at block 305, acquiring a time-series signal comprising multiple samples from one or more electromagnetic energy sensors. For example, multiple samples are acquired from sensor 202 or 204, which may include one or more electromagnetic energy sensors, such as cameras, photodiodes, pyrometers, optical sensors, etc., for measuring the amount of visible or invisible electromagnetic energy emitted from or reflected from the molten pool. The multiple samples correspond to electromagnetic emissions generated by the interaction between a first energy source and powder at a first focal point on the additive material layer, and are generated at a sampling rate greater than or equal to a first sampling rate.

[0045] In some respects, one or more electromagnetic energy sensors are configured to generate time-series signals at high rates, for example, at 10 kHz or higher, 20 kHz or higher, 30 kHz or higher, 40 kHz or higher, 50 kHz or higher, 60 kHz or higher, 70 kHz or higher, 80 kHz or higher, 90 kHz or higher, 100 kHz or higher, or 200 kHz or higher. Figure 4 An exemplary high-rate time-series signal 410 is depicted, which may be generated by one or more electromagnetic energy sensors and obtained by means of implementing method 300.

[0046] Method 300 continues at block 310, decomposing the time-series signal into at least a high-frequency signal variation component. In some respects, the high-frequency signal variation component includes spatial and frequency elements. That is, since the time-series signal is obtained from emission based on the interaction between an energy source and powder at a focal point on the additive layer, the time-series signal includes a spatial component related to the XY plane of the additive layer as the focal point of the interaction moves across the part being constructed.

[0047] The process of decomposing a time-series signal into at least high-frequency signal variation components can be performed using a variety of techniques. In some cases, wavelet packet decomposition can be used to capture a time-series signal and output its high-frequency signal variation components. In certain applications of wavelet packet decomposition, discrete time intervals can be defined such that the high-frequency signal variation components correspond to these defined discrete time intervals. In other cases, continuous wavelet transform can be used to process the time-series signal into its high-frequency signal variation components.

[0048] In some aspects, short-time Fourier transform can be implemented to capture time-series signals and output high-frequency signal variation components. For example, Figure 4 Illustrative low-frequency signal variation component 420 and high-frequency signal variation component 430 are depicted, which can be decomposed individually or separately from the high-rate time series signal 410.

[0049] As used herein, “high frequency” refers to 10 kHz or higher, 20 kHz or higher, 30 kHz or higher, 40 kHz or higher, 50 kHz or higher, 60 kHz or higher, 70 kHz or higher, 80 kHz or higher, 90 kHz or higher, 100 kHz or higher, or 200 kHz or higher. Additive manufacturing anomaly detection techniques target the analysis of high-frequency signal variation components, rather than low-frequency signal variation components, because the problem with conventional monitoring systems is their inability to detect anomalies that are too small to affect the intensity or temperature of the interaction, or anomalies that occur at too short intervals to affect the intensity or temperature, but which could lead to anomalies that could affect the strength, integrity, or structure of the final additively manufactured part. Some example anomalies might include defects of 1 mm or smaller, spatter, small powder droplets, keyhole phenomena (e.g., forming holes), etc.

[0050] Method 300 continues at block 315, calculating one or more spatiotemporal statistics based on the high-frequency signal variation components of the time series signal. In block 315, one or more spatiotemporal statistics include, for example, entropy, energy statistics, percentile comparisons between discrete time intervals, median, standard deviation within discrete time intervals, etc. In some respects, various combinations of multiple spatiotemporal statistics can be calculated for the high-frequency signal variation components. It should be understood that various combinations of spatiotemporal statistics may indicate different anomalies. Therefore, the corresponding spatiotemporal statistics can be calculated according to the type of anomaly to be identified and subsequently processed by a machine learning model.

[0051] In some respects, various spatial and / or temporal relationships between high-frequency signal variation components of a time-series signal can be analyzed to obtain one or more spatiotemporal statistics. For example, spatial relationships can be defined by voxel space, which defines a two-dimensional space of a single layer or a three-dimensional space above and / or below the corresponding layer of a structure. Furthermore, temporal relationships can be defined, for example, by one or more time lengths or sequentially spaced time frames. In some cases, time lengths or time frames can overlap with each other, allowing the obtained statistics to be determined on a rolling basis. In other cases, the obtained statistics can be determined on a rolling basis without temporal or spatial overlap.

[0052] Method 300 continues at block 320, identifying one or more perturbations from one or more spatiotemporal statistics. The presence of one or more perturbations in the signal corresponds to the presence of an anomaly. The process of identifying one or more perturbations from one or more spatiotemporal statistics can be performed by a machine learning model, such as a random forest, a neural network, or other models configured to identify one or more perturbations. In some aspects, the indication of one or more perturbations can be based on the determination or prediction of the machine learning model that one or more perturbations exceed a predetermined threshold corresponding to one or more spatiotemporal statistics. For example, each of the one or more spatiotemporal statistics can have a predetermined threshold that provides an indication of the likelihood of a perturbation when the spatiotemporal statistics exceed the predetermined threshold. In some aspects, the predetermined threshold is built into the model through training, so determining that the spatiotemporal statistics exceed the predetermined threshold is a function of the model, rather than an explicit determination. In some aspects, the predetermined threshold can correspond to control constraints of statistical process control (SPC) of an additive manufacturing machine.

[0053] In some aspects, method 300 continues at block 325, generating an alarm in response to the identification of the presence of one or more disturbances. AM system 100 can represent the alarm in various ways. For example, the alarm can be in the form of an audio message, visual indication, text indication, display prompt, analysis report, etc., provided to the user of the additive manufacturing machine or to another control system to cause one or more further actions to be performed. For example, an alarm may cause the additive manufacturing machine to pause or stop the build process. In some cases, the alarm may include information corresponding to the detected anomaly, such as a prediction of what type of anomaly was detected, such as defects, spatter, powder droplets, surface porosity, etc. In some aspects, the alarm may include spatial information, such as the XY and / or Z position of the anomaly based on the layer number, referred to herein as a layered spatial map. In some aspects, the alarm may include the intensity and / or timestamp corresponding to the anomaly. The alarm and any information defined by the alarm can be further compiled into an analysis report, provided as output to the user. The analysis report may include a heatmap of each layer of the part being built, where the heatmap indicates the probability of an anomaly at a certain location on a particular layer. It should be understood that these are merely some illustrative examples of alarms and how alarms can be represented. The additive manufacturing anomaly detection technology described in this article can generate other forms and types of alerts.

[0054] The functional blocks and / or flowchart elements described herein can be translated into machine-readable instructions or as a computer program product, which, when executed by a computing device, causes the computing device to perform the function of the block. As a non-limiting example, machine-readable instructions can be written using any programming protocol, such as: (ii) descriptive text to be parsed (e.g., Hypertext Markup Language, Extensible Markup Language, etc.), (iii) assembly language, (iv) object code generated from source code by a compiler, (v) source code written using the syntax of any suitable programming language for execution by an interpreter, (v) source code compiled and executed by a real-time compiler, etc. Alternatively, machine-readable instructions can be written in a hardware description language (HDL), such as logic implemented via a field-programmable gate array (FPGA) or an application-specific integrated circuit (ASIC) or its equivalent. Therefore, the functions described herein can be implemented in any conventional computer programming language, as pre-programmed hardware elements, or as a combination of hardware and software components.

[0055] Now for reference Figure 5 The computing system 500 can be deployed on a network. The network may include a wide area network, such as the Internet, a local area network (LAN), a mobile communication network, a public service telephone network (PSTN), and / or other networks. The network can be configured to electronically and / or communicatively connect the computing device 502 and the AMM 501, as per reference. Figure 1 and Figure 2 The AM system 100 is described and depicted.

[0056] Computing device 502 may include a display 502a, a processing unit 502b, and an input device 502c, each of which can be communicatively coupled together and / or connected to a network. Computing device 502 may be a server, personal computer, laptop, tablet, smartphone, handheld device, etc. Computing device 502 can be used by users of the system to provide information to the system. Computing device 502 can access AMM 501 using local or network applications. The system may also include one or more data servers having one or more databases from which computing device 502 and / or AMM 501 can query, retrieve, update, and / or utilize information.

[0057] It is also understandable that while computing device 502 is depicted as a personal computer, this is merely an example. In some respects, any type of computing device (e.g., mobile computing device, personal computer, server, etc.) can be used for any of these components. AMM 501 can be any rapid prototyping, rapid manufacturing, or additive manufacturing equipment, such as binder jet additive manufacturing, fused deposition modeling (FDM), stereolithography (SLA), digital light processing (DLP), selective laser sintering (SLS), selective laser melting (SLM), laminated object fabrication (LOM), electron beam melting (EBM), etc. AMM 501 may include a processor and memory, as well as other electronic components for receiving the computer model of part 503 for printing. The computer model can be converted into a design profile corresponding to part 503 for additive manufacturing and can be uploaded to AMM 501, for example, by computing device 502.

[0058] like Figure 5 As shown, computing device 502 includes processor 510, input / output hardware 512, network interface hardware 514, data storage unit 530, and memory module 520. Memory module 520 may be machine-readable memory (also referred to as non-transient processor-readable memory). Memory module 520 may be configured as volatile and / or non-volatile memory, and therefore may include random access memory (including SRAM, DRAM, and / or other types of random access memory), flash memory, registers, optical disc (CD), digital versatile optical disc (DVD), and / or other types of storage components. Furthermore, memory module 520 may be configured with storage operation logic 522, exception detection logic 524 (e.g., ...), Figure 3 The logic enable method 300 and model logic 526 shown (e.g., logic enabling the machine learning model described herein), as examples, can each be embodied as a computer program, firmware, or hardware. Local interface 540 is also included. Figure 5 It can also be implemented as a bus or other interface to facilitate communication between components of computing device 502.

[0059] Processor 510 may include any processing unit configured to receive and execute programmed instructions (e.g., from data storage unit 530 and / or memory module 520). Instructions may be in the form of a machine-readable instruction set stored in data storage unit 530 and / or memory module 520. Input / output hardware 512 may include a monitor, keyboard, mouse, printer, camera, microphone, speaker, and / or other devices for receiving, transmitting, and / or presenting data. Network interface hardware 514 may include any wired or wireless network hardware, such as a modem, LAN port, Wi-Fi card, WiMax card, mobile communication hardware, and / or other hardware for communicating with other networks and / or devices.

[0060] It should be understood that the data storage component 530 may be located locally or remotely from the computing device 502, and may be configured to store one or more pieces of data for access by the computing device 502 and / or other components. For example... Figure 5 As shown, the data storage unit 530 can store a computer model 532 (e.g., a CAD model) of the part 503 to be manufactured, training data 534 for training a machine learning model, and other data enabling the additive manufacturing anomaly detection techniques described herein. As described herein, the computer model 532 can be a CAD model of the part 503 to be manufactured. The CAD model can define the properties of the green portion of the part used for additive manufacturing, such as defined geometry, material, design tolerances, powder size distribution, type and quantity of binder per layer, sintering profile, coefficient of thermal expansion, etc.

[0061] It should now be understood that the systems and methods described herein use machine learning approaches to quantify the deformation and variability of components, such as binder-sprayed parts, during sintering. In some aspects, the method includes receiving an image of a complex geometry that can be discretized into multiple finite element methods, processing the image of the complex geometry with a machine learning model, predicting the deformation of the complex geometry in response to the sintering process, wherein the machine learning model is trained to predict the deformation of an original geometric specimen with fewer finite element methods than the complex geometry, and generating an image of the predicted deformation of the complex geometry. A surrogate model is trained to predict simple geometries, such that when a complex geometry is input into the model, the trained surrogate model can correlate the learned behavior of the simple part after sintering with the discrete features of the complex geometry, thereby enabling it to predict the deformation of the complex geometry component. Furthermore, the model can predict the average deformation and the variability of the deformation due to the variability of the input dataset (green density, sintering temperature, etc.).

[0062] Other aspects of this disclosure are provided by the subject matter of the following clauses:

[0063] An additive manufacturing anomaly detection method includes: obtaining a time-series signal comprising multiple samples from an electromagnetic energy sensor, wherein the samples correspond to electromagnetic emissions generated by the interaction between a first energy source and powder at a first focal point on an additive material layer, and generating the samples at a sampling rate greater than or equal to a first sampling rate; decomposing the time-series signal into at least high-frequency signal variation components; calculating one or more spatiotemporal statistics based on the high-frequency signal variation components of the time-series signal; identifying one or more disturbances from the one or more spatiotemporal statistics using a machine learning model, wherein the one or more disturbances exceed a predetermined threshold corresponding to the one or more spatiotemporal statistics; and generating an alarm in response to identifying the one or more disturbances.

[0064] According to any one of the preceding clauses, the method of decomposing the time series signal into at least the high-frequency signal variation components includes passing the time series signal through a wavelet packet decomposition process to decompose the time series signal into the high-frequency signal variation components.

[0065] According to any one of the preceding clauses, the method of decomposing the time series signal into at least the high-frequency signal variation components includes performing a short-time Fourier transform on the time series signal to generate the high-frequency signal variation components.

[0066] The method according to any one of the preceding clauses further includes selecting a time period of the time series signal, the time period being a part of the time series signal, and wherein calculating the one or more spatiotemporal statistics includes calculating the one or more spatiotemporal statistics based on the high-frequency signal variation components associated with the part of the time series signal.

[0067] According to any one of the preceding clauses, identifying the one or more disturbances includes feeding the one or more spatiotemporal statistics and the high-frequency signal variation component of the time series signal into the machine learning model, the machine learning model being configured to predict the presence of the one or more disturbances based on the one or more spatiotemporal statistics and the high-frequency signal variation component of the time series signal.

[0068] The method according to any one of the preceding clauses further includes generating an analysis report, the analysis report including a hierarchical spatial map, the hierarchical spatial map including visual indications of the one or more disturbances, wherein the visual indications correspond to at least one of the intensity or location of the one or more disturbances.

[0069] The method according to any one of the preceding clauses, wherein generating the alarm includes causing the additive manufacturing machine to issue an audio or visual instruction.

[0070] The method according to any one of the preceding clauses further includes: depositing the additive material layer on a powder bed of an additive manufacturing machine; and selectively directing energy from the first energy source to the first focal point on the additive material layer.

[0071] According to any one of the preceding clauses, the first sampling rate is a rate greater than or equal to 10 kHz.

[0072] An additive manufacturing system includes: a powder deposition system for depositing an additive material layer on a powder bed of an additive manufacturing machine; a first energy source for selectively directing a first energy beam to a first focal point on the additive material layer; an electromagnetic energy sensor configured to generate a time-series signal comprising a plurality of samples at a sampling rate greater than or equal to a first sampling rate, the plurality of samples corresponding to electromagnetic emissions generated by an interaction between the first energy source and powder at the first focal point on the additive material layer; and a controller operatively coupled to the electromagnetic energy sensor, the controller being configured to: obtain the time-series signal from the electromagnetic energy sensor; decompose the time-series signal into at least high-frequency signal variation components; calculate one or more spatiotemporal statistics based on the high-frequency signal variation components of the time-series signal; identify one or more disturbances from the one or more spatiotemporal statistics, the one or more disturbances exceeding a predetermined threshold corresponding to the one or more spatiotemporal statistics; and generate an alarm in response to identifying the one or more disturbances.

[0073] In the additive manufacturing system according to any one of the preceding clauses, in order to decompose the time series signal into at least high-frequency signal variation components, the controller is further configured to cause the time series signal to undergo a wavelet packet decomposition process.

[0074] In the additive manufacturing system according to any one of the preceding clauses, in order to decompose the time-series signal into at least the high-frequency signal variation components, the controller is further configured to implement the short-time Fourier transform of the time-series signal.

[0075] The additive manufacturing system according to any one of the preceding clauses, wherein the controller is further configured to select a time period of the time series signal, the time period being a portion of the time series signal, and wherein calculating the one or more spatiotemporal statistics includes calculating the one or more spatiotemporal statistics based on the high-frequency signal variation components associated with the portion of the time series signal.

[0076] In the additive manufacturing system according to any one of the preceding clauses, in order to identify the one or more disturbances, the controller is further configured to feed the one or more spatiotemporal statistics and the high-frequency signal variation component of the time series signal to the machine learning model, the machine learning model being configured to predict the presence of the one or more disturbances based on the one or more spatiotemporal statistics and the high-frequency signal variation component of the time series signal.

[0077] The additive manufacturing system according to any one of the preceding clauses, wherein the controller is further configured to generate an analysis report including a layered spatial diagram including visual indications of the one or more disturbances, wherein the visual indications correspond to at least one of the intensity or location of the one or more disturbances.

[0078] In any of the preceding clauses, generating the alarm in the additive manufacturing system comprises causing the additive manufacturing machine to issue an audio or visual instruction.

[0079] The additive manufacturing system according to any one of the preceding clauses, wherein the controller is further configured to: deposit the additive material layer on a powder bed of an additive manufacturing machine; and selectively direct energy from the first energy source to the first focal point on the additive material layer.

[0080] In the additive manufacturing system according to any one of the preceding clauses, the first sampling rate is a rate greater than or equal to 10 kHz.

[0081] An apparatus includes: a controller including one or more processors and one or more memories coupled to the one or more processors, the controller being configured to cause the apparatus to: acquire a time-series signal comprising a plurality of samples from an electromagnetic energy sensor, wherein the samples correspond to electromagnetic emissions generated by an interaction between a first energy source and powder at a first focal point on an additive material layer, and generate the samples at a sampling rate greater than or equal to a first sampling rate; decompose the time-series signal into at least high-frequency signal variation components; calculate one or more spatiotemporal statistics based on the high-frequency signal variation components of the time-series signal; identify one or more disturbances from the one or more spatiotemporal statistics, the one or more disturbances exceeding a predetermined threshold corresponding to the one or more spatiotemporal statistics; and generate an alarm in response to the identification of the one or more disturbances.

[0082] The apparatus according to any one of the preceding clauses, wherein the first sampling rate is a rate greater than or equal to 10 kHz.

[0083] The above description is provided to enable those skilled in the art to practice the various aspects described herein. The examples discussed herein do not limit the scope, applicability, or aspects set forth in the claims. Various modifications to these aspects will be apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects. For example, the function and arrangement of the elements discussed may be altered without departing from the scope of this disclosure. Various procedures or components may be appropriately omitted, substituted, or added to the various examples. For example, the described methods may be performed in a different order than described, and various actions may be added, omitted, or combined. Furthermore, features described with respect to some examples may be combined in some other examples. For example, any number of aspects described herein may be used to implement an apparatus or practice. Moreover, the scope of this disclosure is intended to cover such apparatus or methods practiced using structures, functions, or structures and functions other than those disclosed herein. It should be understood that any aspect of the disclosure herein may be embodied by one or more elements of the claims.

[0084] The various illustrative logic blocks, modules, and circuits described in this disclosure can be implemented or performed using a general-purpose processor, AI processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA) or other programmable logic device (PLD), discrete gate or transistor logic, discrete hardware components, or any combination thereof, designed to perform the functions described herein. The general-purpose processor can be a microprocessor, but can also be any commercially available processor, controller, microcontroller, or state machine. The processor can also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors combined with a DSP core, a SoC, a SiP, or any other such configuration.

[0085] As used herein, the phrase “at least one” in a list of items refers to any combination of these items, including a single member. For example, “at least one of a, b, or c” is intended to cover a, b, c, ab, ac, bc, and abc, as well as any combination of multiples of the same element (e.g., aa, aaa, aab, aac, abb, ac, bb, bbb, bbc, cc, and ccc, or any other order of a, b, and c).

[0086] As used herein, the term "determine" encompasses a wide variety of actions. For example, "determine" may include calculating, processing, deriving, investigating, searching (e.g., looking in a table, database, or other data structure), identifying, etc. Furthermore, "determine" may include receiving (e.g., receiving information), accessing (e.g., accessing data in memory), etc. Additionally, "determine" may include resolving, selecting, picking, establishing, etc.

[0087] As used herein, unless otherwise stated, “connected to” and “connected to” generally include both direct and indirect connections (e.g., including intermediate connection aspects). For example, stating that a processor is connected to memory allows for direct connection or connection via an intermediate aspect (such as a bus).

[0088] The methods disclosed herein include one or more actions for implementing these methods. These actions may be interchanged without departing from the scope of the claims. In other words, the order and / or use of specific actions may be modified without departing from the scope of the claims unless a particular order of actions is specified. Furthermore, the various operations of the above methods can be performed by any suitable means capable of performing the corresponding functions. Such means may include various hardware and / or software components and / or modules, including but not limited to circuits, ASICs, or processors.

[0089] The following claims are not intended to be limited to the aspects shown herein, but should be given the full scope consistent with the language of the claims. Unless otherwise specified, the singular form of an element does not mean "only one," but rather "one or more." Unless otherwise expressly stated, the subsequent use of definite articles (e.g., "the" or "the") and elements (e.g., "processor") does not imply a reference to the singular meaning (e.g., "only one") on the element. For example, unless otherwise specified, references to elements (e.g., "processor," "the processor," etc.) should be understood to refer to one or more elements (e.g., "one or more processors," etc.). The terms "set" and "group" are intended to include one or more elements and may be used interchangeably with "one or more." When referring to one or more elements performing a function (e.g., steps of a method), one element may perform all the functions, or multiple elements may perform these functions together. When more than one element performs a function together, each function does not need to be performed by each of these elements (e.g., different functions may be performed by different elements) and / or each function does not need to be performed by only one element as a whole (e.g., different elements may perform different sub-functions of a function). Similarly, when references are made to one or more elements configured to function another element (e.g., a device), one element may be configured to cause another element to perform all functions, or multiple elements may be collectively configured to cause other elements to perform those functions. Unless otherwise expressly stated, the word "some" refers to one or more. All structural and functional equivalents of the elements relating to the aspects described in this disclosure that are known to or will be known hereafter by those skilled in the art are intended to be covered by the claims. Furthermore, nothing disclosed herein is intended to be made public, regardless of whether it is expressly referenced in the claims.

[0090] It is worth noting that the terms “substantially” and “approximately” are used herein to indicate the degree of uncertainty that may be attributable to any quantitative comparison, value, measurement, or other representation. These terms are also used herein to indicate the extent to which a quantitative representation may differ from the stated reference without altering the fundamental function of the subject matter.

[0091] While specific aspects 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. Furthermore, although various aspects of the claimed subject matter have been described herein, these aspects need not 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 method for detecting anomalies in additive manufacturing, characterized in that, The method includes: A time-series signal comprising multiple samples is obtained from an electromagnetic energy sensor, wherein the samples correspond to electromagnetic emission generated by the interaction between a first energy source and powder at a first focal point on an additive material layer, and the samples are generated at a sampling rate greater than or equal to a first sampling rate. The time series signal is decomposed into at least high-frequency signal variation components; Calculate one or more spatiotemporal statistical values ​​based on the high-frequency signal variation components of the time series signal; Using a machine learning model, one or more perturbations are identified from the one or more spatiotemporal statistics, wherein the one or more perturbations exceed a predetermined threshold corresponding to the one or more spatiotemporal statistics; and An alarm is generated in response to the identification of one or more disturbances.

2. The method according to claim 1, characterized in that, in, Decomposing the time series signal into at least the high-frequency signal variation components includes passing the time series signal through a wavelet packet decomposition process to decompose the time series signal into the high-frequency signal variation components.

3. The method according to claim 1, characterized in that, in, Decomposing the time series signal into at least the high-frequency signal variation components includes performing a short-time Fourier transform on the time series signal to generate the high-frequency signal variation components.

4. The method according to claim 1, characterized in that, The method further includes selecting a time period of the time series signal, the time period being a portion of the time series signal, and wherein calculating the one or more spatiotemporal statistics includes calculating the one or more spatiotemporal statistics based on the high-frequency signal variation components associated with the portion of the time series signal.

5. The method according to claim 1, characterized in that, in, Identifying the one or more disturbances includes feeding the one or more spatiotemporal statistics and the high-frequency signal variation components of the time series signal into the machine learning model, which is configured to predict the presence of the one or more disturbances based on the one or more spatiotemporal statistics and the high-frequency signal variation components of the time series signal.

6. The method according to claim 1, characterized in that, The method further includes generating an analysis report, which includes a hierarchical spatial map that includes visual indications of the one or more disturbances, wherein the visual indications correspond to at least one of the intensity or location of the one or more disturbances.

7. The method according to claim 1, characterized in that, in, Generating the alarm includes causing the additive manufacturing machine to issue an audio or visual instruction.

8. The method according to claim 1, characterized in that, Further includes: The additive material layer is deposited on the powder bed of the additive manufacturing machine; as well as Energy is selectively directed from the first energy source to the first focal point on the additive material layer.

9. The method according to claim 1, characterized in that, in, The first sampling rate is a rate greater than or equal to 10 kHz.

10. An additive manufacturing system, characterized in that, include: A powder deposition system for depositing an additive material layer on a powder bed of an additive manufacturing machine; A first energy source, wherein the first energy source is used to selectively direct a first energy beam to a first focal point on the additive material layer; An electromagnetic energy sensor is configured to generate a time-series signal comprising a plurality of samples at a sampling rate greater than or equal to a first sampling rate, the plurality of samples corresponding to electromagnetic emissions generated by the interaction between the first energy source and the powder at the first focal point on the additive material layer; and A controller, operably coupled to the electromagnetic energy sensor, is configured to: The time-series signal is obtained from the electromagnetic energy sensor; The time series signal is decomposed into at least high-frequency signal variation components; Calculate one or more spatiotemporal statistical values ​​based on the high-frequency signal variation components of the time series signal; A machine learning model is used to identify one or more perturbations from the one or more spatiotemporal statistics, wherein the one or more perturbations exceed a predetermined threshold corresponding to the one or more spatiotemporal statistics; as well as An alarm is generated in response to the identification of one or more disturbances.

Images