Structured Optical Components Quality Monitoring and Usage for Additive Manufacturing

JP2024539526A5Inactive Publication Date: 2024-11-27ADDITIVE MONITORING SYSTEMS LLC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
JP2023526874
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2022-09-02
Filing Date
2022-10-04
Publication Date
2024-11-27
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Current additive manufacturing monitoring methods, such as visual inspection and in-process monitoring systems, are inefficient, costly, require specialized knowledge, and do not provide real-time actionable data, often leading to missed defects and potential printer warranty issues.

Method used

The use of structured optical component quality monitoring systems, like Project Fringe, which employs structured light to detect defects through spatially encoded phase and elevation maps, providing real-time actionable information using digital fringe projection and neural networks to analyze in-situ data.

Benefits of technology

Enables real-time detection of defects in additive manufacturing processes, reducing human subjectivity and data processing challenges, and ensuring consistent printer performance without voiding warranties, while offering micron-level resolution and accurate defect prediction.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 00000000_0000_ABST
    Figure 00000000_0000_ABST
Patent Text Reader

Abstract

Provided herein are systems, methods, and apparatus for structured light part quality monitoring for additive manufacturing. Additive manufacturing (AM) systems, commonly referred to as 3D printers, add layers of build material that are subsequently solidified. This layer addition process can introduce defects into the final part. Generally speaking, structured light part quality monitoring for additive manufacturing comprises a structured light monitoring system for detecting defects in the additive manufacturing process. This Project Fringe uses structured light monitoring measurements and material information to estimate the statistical likelihood of defect occurrence.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates generally to the use of a structured light monitoring system to detect defects in additive manufacturing processes. [Background technology]

[0002] The current status of monitoring additive manufacturing is visual inspection, which involves technicians walking around the manufacturing floor and peering into the sight glasses of 3D printers, looking for visual indicators of defects occurring in the printer's powder bed chamber.

[0003] There are a few commercially available in-situ monitoring systems. The best known in the industry is the Weld Pool Monitoring System produced by EOS, which uses a photodiode in the optical path to measure the "temperature" of the weld pool. User experience has shown that this system generates large data sets in the TB range, making it difficult to extract useful information and unclear how to reduce the data for a "go / no-go" decision by automated means. This system is built into EOS 3D printer systems and will not work with different manufacturers and modules of printers.

[0004] Another in-process monitoring system that can be installed on many different 3D printer brands is the Weld Pool Monitoring System from SigmaLabs. This system monitors the weld pool during the 3D printing process, but requires the addition of a built-in sensor to the printer's beamline, which voids the warranty of the sensor. This type of modification prevents widespread use of this type of sensor, and the analysis to make a go-no-go decision can be swamped with huge amounts of data that are too large to process in real time.

[0005] The disadvantages of using visual inspection to assess health (or detect defects) are that 1) it relies on subjective human opinion, 2) it can only be applied to the current layer, 3) it requires costly labor to observe the build, and 4) defects may be missed when a technician is not observing the process.

[0006] The drawbacks of using current in-process monitoring melt pool analysis techniques are: 1) they are cost prohibitive, 2) they do not provide actionable “go / no-go” information in real-time, 3) they require specialized knowledge, 4) they require analysis of large data sets, 5) they are not real-time, and 6) they can void the printer warranty in the case of non-OEMs.

[0007] The present invention seeks to solve these and other problems. Summary of the Invention

[0008] SUMMARY OF THE DISCLOSURE Provided herein are systems, methods, and apparatus for structured optical component quality monitoring for additive manufacturing.

[0009] The methods, systems, and apparatus will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the methods, apparatus, and systems. The advantages of the methods, apparatus, and systems will be realized and attained by means of the elements and combinations particularly pointed out in the appended claims. It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the methods, apparatus, and systems as claimed.

[0010] Thus, it is the object of the present invention not to encompass in the present invention any known products, processes for making the products, or methods for using the products, and as a result, the applicant reserves the right and hereby discloses a disclaimer of the known products, processes, or methods. It is noted that the present invention does not intend to encompass within its scope any products, processes for making the products, or methods for using the products that do not meet the description and enablement requirements of the USPTO (Article 35 USC 112, first paragraph) or the EPO (Article 83 EPC), and as a result, the applicant reserves the right and hereby discloses a disclaimer of the aforementioned products, processes for making the products, or methods for using the products. Compliance with Article 53(c) of the EPC and Rules 28(b) and (c) of the EPC may be advantageous in the practice of the present invention. All rights to expressly disclaim any embodiment that is the subject of the applicant's granted patent(s), in this line, or in any other line, or in any prior application of a third party, are expressly reserved. Nothing herein should be construed as a promise.

[0011] In the accompanying drawings, like elements are identified with like numerals within several preferred embodiments of the present invention. [Brief description of the drawings]

[0012] [Figure 1] 1 is an example of an additive manufacturing system according to some illustrative embodiments. [Diagram 2] FIG. 13 is a schematic diagram showing a map of the difference between the unmelted powder area and the melted surface area. [Diagram 3] FIG. 13 is a schematic diagram of a difference map of a newly deposited powder layer and a previously melted layer. [Figure 4A-4F] 1 is height measurement data showing the shrinkage of powder height after melting. [Diagram 5] 13 is a colored 3D point cloud (Process L-PBF) graph showing a typical elevation map difference, illustrating powder surface elevation minus molten surface elevation, as well as the surface texture and elevation variation characteristics of the unmolten powder bed and molten regions. [Figure 6]3D colored point cloud (Process L-PBF) graph showing typical elevation map difference: powder surface elevation - molten surface elevation, general features of unmelted powder bed, molten areas showing anomalies / defects including over melting, thermal protrusion, recoater vibration / hopping, powder bed thickness variation. [Figure 7] 13 is a colored 3D point cloud (Process L-PBF) graph showing typical elevation map difference, and melt surface elevation minus next powder surface elevation; common features observed include powder bed thickness over both the part (powder over melted area) and unmelted powder bed (powder over powder); anomalies / defects observed include spatter and debris, layer thickness variations. [Figure 8] Grayscale 3D point cloud (Process L-PBF) graph showing typical elevation map difference, post-melt elevation map - next powder layer; common features observed include surface morphology of powder bed and fused area; anomalies / defects observed include recoater streaks. [Figure 9] Grayscale 3D point cloud (Process L-PBF) graph showing the elevation map difference based on the analysis, powder layer - next post-melt elevation map, common features observed include surface morphology of the powder bed and fused area, and anomalies / defects observed include recoater streaks. [Figure 10] 3D point cloud (Process L-PBF) graph of greyscale phase showing the phase map difference based on the analysis, post melt phase map (layer 56) - next powder layer phase map (layer 57), common features observed include phase difference between the melt layer and the powder layer, and anomalies / defects observed include powder pits and bumps. [Figure 11] 3D point cloud (Process L-PBF) graph of greyscale phase showing the phase map difference based on the analysis, powder layer phase map (layer 55) - post-melt phase map (layer 56), common features observed include phase difference between the powder layer and the next melt layer, anomalies / defects observed include powder dents and bumps, detected anomalies. [Figure 12]Colored 3D point cloud (Process L-PBF) graph showing typical elevation maps from fringe projections taken after powder spreading and before melting; common features observed include powder streaks, powder skew, and non-uniform powder distribution. [Figure 13] Colored 3D point cloud (Process L-PBF) graph showing typical elevation maps from fringe projections taken after fusion, where common features include powder streaks, powder skew, non-uniform powder distribution, thermal deformation, molten areas, and extrusions. [Figure 14A-14B] 14A and 14B are graphs showing typical elevation maps from fringe projections of a powder bed down to -50 microns, where the general features include 3D measurements of the powder layer at bed level -50 microns, FIG. 14A shows an elevation map of the powder bed, and FIG. 14B shows a cross section of the elevation map (Process L-PBF). [Figure 15A-15B] 15A and 15B are graphs showing typical elevation maps from fringe projections of a powder bed elevated to +50 microns, where FIG. 15A shows the elevation map of the powder bed and FIG. 15B shows a cross section of the elevation map (Process L-PBF), and observed features include 3D measurements of the powder layer at bed level +50 microns. [Figure 16A-16B] FIG. 16A is a graph showing a powder bed elevation map of a visual image of the binder jetting process after binder deposition, and FIG. 16B is a graph showing a cross section of the elevation map (process binder jetting), a typical phase map from a fringe projection after binder deposition, where observed features include recoater roller powder anomalies, binder powder interactions, binder streaks, ridges, and surface depressions. [Figure 17] 1 is a schematic flow chart showing the training of neural networks and Gaussian process regression on elevation maps. Training the GPR Model. To construct the training data, the in-situ data is aligned with the grayscale CT images to generate "input-output pairs". To avoid significant computational overhead, this training data is then split among an ensemble of K GPR models. [Figure 18]1 is a schematic flow chart showing the operation of neural networks and Gaussian process regression on elevation maps and features including predictions using GPR models. In-situ measurements are made during build and fed into an ensemble of GPR models. The predictions made by each GPR model are then combined to achieve a greyscale value to form an AI-generated porosity predictor. [Figure 19] CT scans for ground truth validation of printed samples, including training data collection examples for Gaussian process regression and neural network training; features include low density areas, potential pore locations. [Figure 20] CT scans for ground truth validation of printed samples, including training data collection examples for Gaussian process regression and neural network training; features include low density areas, potential pore locations. [Figure 21] CT scans for ground truth validation of printed samples, including example training data collection for Gaussian process regression and neural network training; features include low density areas, potential pore locations, and AI / ML identified low density areas. [Figure 22] CT scan ground truth validation of printed samples including example training data collection for Gaussian process regression and neural network training. Features include low density areas, potential pore locations, and AI / ML identified low density areas. [Figure 23A-23B] FIG. 23A is a graph showing printing anomalies observed using advanced mapping (difference between powder and melt) of layer 567, and FIG. 23B is a graph showing in-situ data and AI / ML and Gaussian process regression prediction testing of low density areas, low density locations and / or pore locations identified with AI / ML of slice 567, features include low density areas, potential pore locations, low density areas identified with AI / ML. [Fig. 24A-24B]FIG. 24A is a graph of in situ data of low density locations and / or pore locations showing printing anomalies observed using advanced mapping (difference between powder and melt) of layer 567, and FIG. 24B is a graph of AI / ML and Gaussian process regression prediction tests of low density locations and / or pore locations showing low density areas identified with AI / ML of slice 567, where features include low density areas, potential pore locations, and low density areas identified with AI / ML. [Diagram 25] FIG. 1 is a system diagram of a surveillance system including hardware / software elements. [Figure 26] 1 is a photograph of the monitoring system attached to a printer (Laser Powder Bed Fusion) showing the system configuration. [Figure 27] Photograph showing streak patterns on a powder bed, features include powder areas, streak patterns. [Figure 28] Photograph showing the streak pattern on the powder bed (2), features include molten areas, powder areas, and the streak pattern. [Figure 29A-29B] Figure 29A is a photograph showing the camera system. The camera is attached to an EOS M 290 powder bed fusion 3D printer. Figure 29B is a photograph showing the projector system. The projector is attached to an EOS M 290 powder bed fusion 3D printer. [Diagram 30] Photograph showing the camera system. The camera is mounted on an exone innovent binder jetting 3D printer. The projector is mounted on an exone innovent binder jetting 3D printer. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] The foregoing and other features and advantages of the present invention will become apparent from the following detailed description of illustrative embodiments, which is to be read in conjunction with the accompanying drawings, in which the detailed description and drawings are merely illustrative of the invention rather than limiting, the scope of the invention being defined by the appended claims and equivalents thereof.

[0014]

[0023] An embodiment of the present invention is described herein with reference to the drawings, in which like reference numerals refer to like elements throughout. The terms used in the description set forth herein are not intended to be interpreted as limiting or restrictive, as they are merely utilized in conjunction with a detailed description of certain embodiments of the present invention. Furthermore, an embodiment of the present invention may include several novel features, no single one of which is solely responsible for its desirable properties or is essential to the practice of the invention described herein.

[0015] The words proximal and distal are applied herein to indicate particular ends of components of the instruments described herein. The proximal end refers to the end of the instrument that is closer to an operator of the instrument when the instrument is in use. The distal end refers to the end of the component that is further from the operator and extends toward the surgical area of ​​the patient and / or implant.

[0016] The use of the term "solidification process" or "solidified layer" refers to the process or result of deformation of a metal or polymer powder melting and bonding to a part of the 3D printer or the part being printed. This can occur through a thermal process, such as melting or sintering, or through a chemical bonding process.

[0017] The use of the terms "a" and "an" and "the" and similar designations in the context of describing the present invention should be construed to cover both the singular and the plural, unless otherwise indicated herein or clearly contradicted by context. It will be further understood that the terms "comprises," "comprising," "including," and / or "comprising," as used herein, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0018] The recitation of ranges of values ​​herein is intended to serve merely as a shorthand method of referring individually to each separate value within the range, unless otherwise indicated herein, and each separate value is incorporated herein as if it were individually set forth herein. The word "about", when accompanied by a numerical value, should be construed to indicate a deviation of no more than 10% from the stated numerical value. Any examples or exemplary language provided herein (such as "eg" or "such as") are intended merely to better clarify the invention and do not pose a limitation on the scope of the invention unless otherwise claimed. No language in the specification should be construed as indicating any non-claimed element essential to the practice of the invention.

[0019] References to "one embodiment," "an embodiment," "an exemplary embodiment," "various embodiments," etc. may indicate that the embodiment or embodiments of the invention so described may include a particular feature, structure, or characteristic, but not all embodiments necessarily include that particular feature, structure, or characteristic. Moreover, repeated use of the phrases "in one embodiment" or "in an exemplary embodiment" may refer to the same embodiment, but do not necessarily refer to the same embodiment.

[0020] As used herein, the term "method" refers to ways, means, techniques, and procedures for accomplishing a given task, including, but not limited to, ways, means, techniques, and procedures known to practitioners of the chemical, pharmacological, biological, biochemical, and medical arts, or readily developed by such practitioners from known ways, means, techniques, and procedures. Unless otherwise expressly stated, no method or embodiment described herein is intended to be construed as requiring that its steps be performed in a particular order. Thus, if a method claim does not specifically state in the claims or description that the steps are limited to a particular order, no order is intended to be implied in any respect. This preserves all possible implicit bases of interpretation, including obvious meaning derived from the organization or operational flow of the steps, grammatical organization or punctuation, or questions of logic regarding the number or type of embodiments described in the specification.

[0021] explanation Generally speaking, structured light part quality monitoring for additive manufacturing comprises a structured light monitoring system to detect defects in additive manufacturing processes. This Project Fringe uses structured light monitoring measurements and material information to estimate the statistical likelihood of defect occurrence. Additive manufacturing (AM) systems, commonly referred to as 3D printers, add layers of build material, which are subsequently solidified. This layer-adding process can introduce defects into the final part.

[0022] Project Fringe spatially encodes unique phase values ​​of unidirectional data for each measurement surface. This is accomplished by projecting a predetermined pattern onto the measurement surface, which can be a build plate, powder layer, molten layer, bond layer, or sintered layer. Using the phase values ​​encoded on the measurement surface, process anomalies during the printing process can be detected. These anomalies during the printing process can include, but are not limited to, recoater cross-sectional information, build surface skew, plate skew, build plate flatness, build plate vibration, build plate vertical accuracy, powder coating thickness, powder coating roughness, powder coating streaks, powder coating hopping, powder coating underfeed, powder coating clumps, powder coating agglomeration, changes in powder coating after melting or solidification or bonding, non-uniform powder particles, thermal deformation, overmelting, undermelting, component extrusion, absolute layer height from reference, material collapse, distortion of support structures, total distortion, surface distortion, particle spitting, particle aggregation, particle agglomeration, uneven melting, uneven sintering, laser spacing effects, laser spot size, laser power, binder penetration effects, binder powder interactions, binder deposit high solidification, or any other physical phenomenon that shifts the encoded spatial phase from a measured or mathematical reference plane.

[0023] The features appear in the spatially encoded phase data or can be extracted using matched filters. These features can be extracted using kernel convolution techniques, absolute phase difference measurements, separable convolution, dilated convolution, atrous convolution, deformable convolution, demodulation, Fourier transform, Hilbert transform, Gaussian filtering, flat box filtering, Sobel filters, Sobel operators, Sobel-Fieldman operators, Prewitt operators, Laplace operators.

[0024] In addition to generating spatially phase encoded surface maps, Project Fringe can also generate elevation maps. Project Fringe generates elevation maps of the powder bed chamber that record the elevation of each deposited powder layer and / or solidified layer, classifies these elevation maps as "in spec" or "out of spec," and alerts operators / technicians / engineers of these elevation anomalies.

[0025] Project Fringe comprises modular hardware and software systems that operably couple to many types of 3D printers (additive manufacturing printers) to provide real-time measurements of exposed surfaces of parts and feedstocks. These advanced measurements, apart from phase measurements, can be used to detect process anomalies during the printing process. These anomalies during the printing process can include, but are not limited to, recoater cross-sectional information, build surface skew, plate skew, build plate flatness, build plate vibration, build plate vertical accuracy, powder coating thickness, powder coating roughness, powder coating streaks, powder coating hopping, powder coating underfeed, powder coating clumps, powder coating agglomeration, changes in powder coating after melting or solidification or bonding, non-uniform powder particles, thermal deformation, overmelting, undermelting, component extrusion, absolute layer height from reference, material collapse, distortion of support structures, total distortion, surface distortion, particle spitting, particle aggregation, particle agglomeration, uneven melting, uneven sintering, laser spacing effects, laser spot size, laser power, binder penetration effects, binder powder interactions, binder deposit high solidification, or any other physical phenomenon that shifts the encoded spatial phase from a measured or mathematical reference plane.

[0026] The features appear in the elevation data or can be extracted using matched filters. These features can be extracted using kernel convolution techniques, absolute elevation difference measurements, separable convolution, dilated convolution, atrous convolution, deformable convolution, demodulation, Fourier transform, Hilbert transform, Gaussian filtering, flat box filtering, Sobel filters, Sobel operators, Sobel-Fieldman operators, Prewitt operators, Laplace operators.

[0027] Project Fringe generates a unique spatially encoded phase value for each pixel on the measurement area, and areas of non-uniform phase constitute process anomalies during 3D printing. Project Fringe also creates an elevation map. The elevation map of the exposed surface is used to detect 1) feedstock layers (dimples or lumps) that contribute to final part failures and 2) elevation anomalies in the newly solidified layers that contribute to final part defects and printer damage.

[0028] The hardware system comprises a structured light system, specifically a digital fringe projection (DFP). The DFP consists of a projector, a camera, and a computer connected to each by timing, triggering, and data acquisition circuitry to control each and perform mathematical operations on the images captured by the camera. The DFP projects a pattern onto both the part surface and the feedstock surface, records the deformation of the pattern, includes a spatially encoded phase on the exposed surface, and separately performs 3D measurements by calculating an elevation map that represents the 3D surface of the measurement. This elevation map, triggered before and after the laser solidification process, can reveal many different types of in-situ defects that may ultimately cause the entire part to be rejected at the end stage. These in-situ defects include, but are not limited to, warpage, overmelt, undermelt, thermal distortion, poor fusion bonding, recoater blade defects, build plate skew, under-powder feeding, non-uniform powder coating, recoater blade hopping, powder streaks, powder pits, and part extrusion. The Project Fringe system estimates the impact of in-situ detected defects on the quality of the final part and makes decisions based on this estimate.

[0029] Providing these elevation maps that are sensitive to these in-situ defects is valuable to anyone involved in building 3D printed parts. However, further added value from this measurement is gained by correlation of these in-situ defects with defects in the final part. Included in the IP the inventors are filing is the practice of linking the in-situ defects to voids / pores in the final part. This is accomplished in two ways: first, by linking the in-situ elevation maps with CT and x-ray scans of the final part to establish size thresholds for in-situ defect sizes that, if exceeded, will cause void formation in the final part, and second, by using physical models to establish these size limits in the melting or sintering or bonding process. Each or both of these methods can be employed in the present invention.

[0030] In one embodiment, these detected defects are compared to a library of previous 3D prints in which similar defects were recorded and which resulted in defects in the final part to provide an estimate of the likelihood of a defect being present in the final part. Correlation of in-situ and ex-situ defects may not be required in some embodiments.

[0031] Project Fringe, with its Digital Fringe Projection (DFP) monitoring system, can be used for Powder Bed Fusion (PBF), Binder Jetting (BJ), and Selective Laser Sintering (SLS) polymer additive manufacturing (AM). Although some of the examples described herein refer to the use of Project Fringe to quantify the uncertainty of DFP advanced measurements, Project Fringe can also be used to provide an indication of the uncertainty associated with other types of optical or light-based measurement techniques, including digital image correlation techniques, contact scanner techniques, coherent light imaging techniques, and the like.

[0032] In another embodiment, Project Fringe geometrically measures layer height before and after solidification to detect warpage, powder bed defects, fusion bond failures, and delamination. A pattern is projected onto the part surface before and after laser processing and a 3D height map is calculated based on the pattern deformation. Project Fringe provides modular spatial measurement density, where every pixel is an height measurement, providing micron-level resolution of height features.

[0033] Project Fringe records elevation maps of each layer to provide actionable information in real time and detect defects, without the need for extensive data storage or analysis. Project Fringe measures different features (elevation maps of the powder bed and molten layer). Project Fringe provides real time part quality estimation (go / no-go decisions).

[0034] Project Fringe calculates the accuracy of each measurement point, estimates the presence of internal defects based on in-situ topography using previous data on defects in final parts, operably couples to existing printers via a mounting mechanism, uses neutral density filters on the camera and creates a phase map after using algorithms to filter the image, which in one embodiment is then converted to a topography and estimates the density of the part based on the phase or topography information.

[0035] In one embodiment, Project Fringe is mounted inside or outside the 3D printer chamber 110 as shown in FIG. 1. A computer will be used to calibrate the measurement system. The monitoring system computer triggers the projection and camera image capture and processes the images. There can be any number of cameras used for fringe projection measurements. The projectors can provide different projections, which can be simultaneously superimposed with different fringe wavelengths. Also, the wavelength of the projected light can be varied during or before projection. The projectors can be the light source and can be located at specific locations both inside and outside the 3D printer chamber. Additionally, there may be multiple cameras, projectors, and / or interferometers. The cameras can be image sensors, imaging structured light such as fringes projected onto the surface of the object being measured. The camera position can be kept constant at the observation angle. The pixel patch spacing can be on the order of microns. The projector output trigger function triggers the camera image acquisition to capture fringe images at regular intervals, eliminating projector "fringing" that occurs without synchronization. To provide user interaction, the subject matter described herein may be implemented in a computer having a display device, such as, for example, a cathode ray tube (CRT) or liquid crystal display (LCD) monitor, for displaying information to the user, as well as a keyboard and a pointing device, such as, for example, a mouse or trackball, through which the user can provide input to the computer. Other types of devices, such as mobile phones, or portable computers, or tablets, may also be used to provide user interaction. For example, feedback provided to the user may be any form of sensory feedback, such as, for example, visual feedback, auditory feedback, or haptic feedback, and input from the user may be received in any form, including, but not limited to, acoustic, speech, or haptic input.

[0036] DFP altitude measurements are made by projecting a pattern onto a flat reference surface (either physical or mathematical, further indicated by the subscript "r"), then placing an object in situ and recording how the projected pattern deforms due to the object's shape (further indicated by "o"). Calculation of the resulting deformation of the fringe pattern produces a phase map, which is then converted to an altitude map by a calibration routine. During in-situ measurements, inside the PBF machine, the reference surface can be taken as a bare base plate or a base plate with a uniform initial powder coating. If the DFP geometry is assumed to be constant, a reference phase map can be mathematically generated, allowing differential altitude measurements independent of any altitude profile of the reference surface.

[0037] Working Example The following examples are presented to provide those skilled in the art with a complete disclosure and description of how the compounds, compositions, articles, devices, and / or methods claimed herein are made and evaluated, and are intended to be purely illustrative of the invention and are not intended to limit the scope of what the inventors regard as their invention. However, those skilled in the art should understand in light of this disclosure that many modifications can be made to the specific embodiments disclosed and still obtain the same or similar results without departing from the spirit and scope of the invention. Efforts have been made to ensure accuracy with respect to numbers (e.g., amounts, temperature, etc.), but some errors and deviations should be accounted for. Unless otherwise indicated, parts are parts by weight, temperature is ° C. or at ambient temperature, and pressure is at or near atmospheric pressure.

[0038] In polymer powder bed fusion (PBF) additive manufacturing, of which selective laser sintering is an example, shrinkage is easily visible to the naked eye with good shrinkage being an indicator of good sintering and a stable process. In SLS, the powder bed is preheated to about 80% of the material's melting temperature. The laser power adds the remaining ~20% of the energy to cause sintering.

[0039] Powder recycling is an important part of polymer PBF processes such as SLS. In SLS, powder recovered from a build after the parts have been removed is often refreshed by mixing it with new unused powder (virgin powder) and discarding some of the used powder as waste. This is done to ensure that the powder used in the build has the correct properties to ensure a stable build process, but it is because the powder undergoes multiple heatings during multiple builds that result in changes in crystallinity and make the powder more difficult to melt or sinter. If virgin powder is not mixed into the recycle batch, users are forced to increase the build chamber preheat temperature or increase the laser power to allow for good melting and good part production. Ultimately, if virgin powder is not added to the recycle material, the batch becomes unprocessable and must be discarded as waste.

[0040] Since shrinkage is an important indicator of good melting, if it can be measured to a known level of uncertainty it can be used to allow the user to monitor build quality and / or adjust machine settings via a control feedback loop, or make educated decisions about discarding or mixing reused powder. The roughness of the top surface of the solidified material can also provide an important indicator of build quality.

[0041] Laser and E-beam PBF AM of metals is similar in its operation to polymer PBF systems, but does not require the same quantities if there is powder recycling. This is because, in contrast to polymer PBFs, there is little change in powder properties over time and most of the energy to melt the powder comes from the E-beam or laser rather than from pre-heating the powder.

[0042] Inkjet bonding systems rely on the application of a drop of binder to the part surface and involve several additional mechanisms. As the binder is applied, it soaks into the powder. As the binder soaks into the powder, it pulls the loose powder bed together creating shrinkage. Data showing the binder and powder shrinkage are shown in Figures 4A-4F.

[0043] In an inkjet bonding system, if there is too much binder, a meniscus will form on the powder surface. In an inkjet bonding system, if there is too little binder, the surface will not shrink properly. Thus, in an inkjet bonding system, proper wetting and ultimately shrinkage will have a characteristic optimum drop onto the surface. The methods and systems described herein can be used to detect and optimize this shrinkage and control the performance of an inkjet printer.

[0044] As shown in FIG. 2, a method 200 for detecting shrinkage and estimating print quality in-process includes measuring the height difference 240 between the deposited layer 210, the unmelted material 220, and the solidified / melted bond material 230. A sensing device 250 measures the height difference 240 and, according to one embodiment, is comprised of a camera and lighting system. In one embodiment, the height difference is about 1 micron to about 10 microns, alternatively about 2 microns to about 8 microns, alternatively about 3 microns to about 6 microns, alternatively about 4 microns. In another embodiment, the sensing device is a contact imaging device or a laser line profilometer. The amount that the layer is reduced upon melting is shown in FIG. 2. In one embodiment, the amount of layer reduction is about 10 microns to about 80 microns, alternatively about 15 microns to about 70 microns, alternatively about 20 microns to about 60 microns.

[0045] In one embodiment, the method and system includes assessing the integrity of a powder based on how much the powder loses depth during powder bed fusion, electron beam fusion, or polymer sintering.

[0046] In one embodiment, the method and system includes a feedback loop for the chamber temperature to control melting of the recycled powder from shrinkage measurements.The temperature sensor may detect the chamber ambient temperature, according to one embodiment.

[0047] In one embodiment, the method and system includes a feedback loop for the laser power to control the melting of the recycled powder. In one embodiment, the laser power is measured or sensed from the laser system.

[0048] In one embodiment, the method and system includes observing a decrease in binder jetting power altitude of the powder layer after the binder is added. In one embodiment, the method and system includes observing an increase in altitude between the molten / solid layer (and bonded layer) and the next unmolten powder layer.

[0049] In one embodiment, the method and system includes estimating post-build part quality based on measuring the height difference between the deposited material, the unmelted material, or the solidified / molten bonded material. The method and system includes measuring the height difference between the deposited unmelted material and the solid layer below, and measuring the height difference between the unmelted material and the solidified-sintered / molten or bonded material. The method and system further includes estimating porosity, grain structure, and mechanical properties, and inferring part quality using these estimates. In one embodiment, the percent density ranges from about 99% to about 100%, and the porosity ranges from about 0% to about 1%. In one embodiment, the grain structure varies from martensite to austenite, and the grain size distribution ranges from about 0.5 μm to about 500 μm. The mechanical properties include tensile strength, ultimate yield strength, fatigue life, elongation, or cracking. The binder jet density varies between about 75% to about 95%.

[0050] In one embodiment, estimating the porosity, grain structure, and mechanical properties is accomplished using a physics-based computer model, hi one embodiment, estimating the porosity, grain structure, and mechanical properties is performed using a library of prior measurements and / or neural networks, Gaussian process regression, convolution kernels, and the like.

[0051] Neural networks (NNs), also called artificial neural networks (ANNs) or simulated neural networks (SNNs) in the case of artificial neurons, are interconnected groups of natural or artificial neurons that use mathematical or computational models for information processing based on connectionistic computational techniques. In most cases, ANNs are adaptive systems that change their structure based on external or internal information flowing through the network. In more practical terms, neural networks are nonlinear statistical data modeling or decision-making tools. They can be used to model complex relationships between inputs and outputs or to find patterns in data.

[0052] Gaussian process regression is non-parametric (i.e., not restricted by the form of the function), so rather than calculating a probability distribution over the parameters of a particular function, GPR calculates a probability distribution over all allowable functions that fit the data.

[0053] In a convolutional kernel or convolutional neural network, the kernel is nothing more than a filter used to extract features from an image. The kernel is a matrix that moves over the input data and performs dot products with subregions of the input data, and gets the output as a matrix of dot products.

[0054] In one embodiment, the method and system includes using the altimetry to estimate the amount of powder lost as ejection and evaporation during solidification, and using the altimetry to optimize performance. In one embodiment, an amount of powder is lost in metal additive manufacturing.

[0055] 3, a method 300 for estimating recoating fill fraction includes measuring the elevation of deposited and unmelted material, measuring the difference between a previously melted layer 310 and a newly deposited layer 320, and determining how well the next layer will fill via an elevation map 330. In one embodiment, the fill is measured by percent loss and elevation shrinkage.

[0056] Information on neural networks and physical models for defect estimation Proven performance in AM. The GPR model has recently been shown to be successful in predicting porosity in laser powder bed fusion bonded parts using in situ measurements from a Renishaw AM 500M machine

[10] . A GPR model was trained to predict porosity using in situ features. Tests were performed over a wide density range, achieving the results shown in Figure 17, reporting an RMS error of less than 4%.

[0057] Uncertainty quantification. AI must provide some form of uncertainty quantification. The advantage of GPR models over neural networks is that they are probabilistic and therefore inherently provide confidence limits on their predictions without significant additional computation; the results shown in Figures 17-18 show the uncertainty bounds for GPR models.

[0058] Scalability. In situ AM data (and CT images) can be of significant size (often GB per build), and AI models must scale well with the size of the training data. GPR models include a product-of-experts approach, which allows GPR models to scale to large datasets without specialized computational architectures

[11] . Rather than training a single GPR, which would be costly for large datasets, the training data is split evenly among K different "AI experts" to predict part porosity from the in situ data, and the final prediction is a weighted combination of the predictions made by each AI expert. Importantly, such an approach does not require all experts to be in memory simultaneously, avoiding RAM limitations that are a bottleneck for many AI approaches.

[0059] Out-of-data performance. After model training, builds may be made in which phenomena occur that did not occur during the generation of the training data. In such cases, the AI ​​model is being applied to new scenarios not represented by the training data, and GPR models have the advantage that they fail gracefully when applied out-of-data, and prediction confidence limits can be significantly expanded. This is preferable to other methods, including neural networks, which tend to provide confidently erroneous estimates when applied outside of their training data

[12] .

[0060] Figure 18 (see PowerPoint) shows the process required to initialize a GPR model. A build is performed using a Renishaw AM400 while in situ measurements (i.e. Project Fringe elevation map data, visual images) are collected in parallel with data from the press log of the machine. After the build, a CT scan of the build components is performed. The in situ data is then aligned with the CT data using least squares matching for the 3D data. The CT results are expected to be at higher resolution than the in situ measurements, and therefore each in situ measurement is associated with an average value of the surrounding CT data. This process generates a set of input-output pairs where the input is the in situ measurement and the output is the corresponding porosity measurement obtained from the CT results. This data is then split between several GPR "expert" models to avoid the large computational overhead of training a single GPR model on a large data set.

[0061] A gradient-based algorithm is trained to obtain maximum likelihood parameter estimates. The data is divided into k equal-sized sets, so that k-1 data sets can be used to train the model, leaving the remaining set for validation. The model is tested in k iterations, with a different combination of test and training data sets used in each iteration. The performance of the final model is determined by taking the average of the ensemble of tests, a process known as k-fold cross-validation. The performance of the model was measured using the area under the curve technique, i.e., the area under the receiver operating characteristic (ROC) curve obtained when varying the model's threshold decision value (the value above which the part is determined to contain porosity based on the model's results). The ROC curve was used to analyze the performance of the model over the entire range of possible thresholds. These tests are repeated k times, and for each test, the roles of the training and test data are swapped after the k-fold cross-validation procedure.

[0062] FIG. 5 is a colored 3D point cloud (Process L-PBF) graph showing a typical elevation map difference, showing powder surface elevation minus molten surface elevation, as well as the surface texture and elevation variation characteristics of the unmolten powder bed and molten regions.

[0063] Figure 6 is a colored 3D point cloud (Process L-PBF) graph showing a typical elevation map difference: powder surface elevation - molten surface elevation, general features of unmelted powder bed, molten area showing anomalies / defects including over melting, edge protrusion, recoater vibration / hopping, powder bed thickness variation.

[0064] Figure 7 is a colored 3D point cloud (Process L-PBF) graph showing a typical elevation map difference and melt surface elevation minus the next powder surface elevation, common features observed include powder bed thickness both above the part (powder above the melted area) and above the unmelted powder bed (powder above powder), anomalies / defects observed include spatter and debris, layer thickness variations.

[0065] Figure 8 is a grayscale 3D point cloud (Process L-PBF) graph showing a typical elevation map difference, post-melt elevation map - next powder layer, where common features observed include surface morphology of the powder bed and fused portion, and anomalies / defects observed include recoater streaks.

[0066] FIG. 9 is a grayscale 3D point cloud (Process L-PBF) graph showing the height map difference based on the analysis, powder layer-next post-melt height map, where common features observed include surface morphology of the powder bed and fused parts, and anomalies / defects observed include recoater streaks.

[0067] FIG. 10 is a grayscale phase 3D point cloud (Process L-PBF) graph showing the phase map difference based on the analysis, post melt phase map (layer 56) minus next powder layer phase map (layer 57), where common features observed include phase difference between the melt layer and the powder layer, and anomalies / defects observed include powder pits and bumps.

[0068] FIG. 11 is a 3D point cloud (Process L-PBF) graph of grayscale phase showing the phase map difference, powder layer phase map (layer 55) - post-melt phase map (layer 56) based on the analysis, common features observed include phase difference between the powder layer and the next melt layer, and anomalies / defects observed include powder pits and bumps, detection anomalies.

[0069] Figure 12 is a colored 3D point cloud (Process L-PBF) graph showing a typical elevation map from a fringe projection taken after powder spreading and before melting, where common features observed include powder streaks, powder skew, and non-uniform powder distribution.

[0070] Figure 13 is a colored 3D point cloud (Process L-PBF) graph showing a typical elevation map from a fringe projection taken after fusion, where common features include powder streaks, powder skew, non-uniform powder distribution, thermal deformation, molten areas, and extrusions.

[0071] 14A-14B are graphs showing typical elevation maps from fringe projections of a powder bed down to -50 microns, where the general features include 3D measurements of the powder layer at bed level -50 microns, where FIG. 14A shows the elevation map of the powder bed, and FIG. 14B shows a cross section of the elevation map (Process L-PBF).

[0072] 15A-15B are graphs showing typical elevation maps from fringe projections of a powder bed elevated to +50 microns, where FIG. 15A shows the elevation map of the powder bed and FIG. 15B shows a cross section of the elevation map (Process L-PBF), and the observed features include 3D measurements of the powder layer at bed level +50 microns.

[0073] FIG. 16A is a graph showing a powder bed elevation map of a visual image of the binder jetting process after binder deposition, and FIG. 16B is a graph showing a cross section of the elevation map (process binder jetting), a typical phase map from a fringe projection after binder deposition, where observed features include recoater roller powder anomalies, binder powder interactions, binder streaks, ridges, and surface depressions.

[0074] FIG. 17 is a schematic flow chart showing the training of neural networks and Gaussian process regression on elevation maps. Training the GPR Model. To construct the training data, the in-situ data is aligned with the grayscale CT images to generate "input-output pairs". To avoid significant computational overhead, this training data is then split among an ensemble of K GPR models.

[0075] FIG. 18 is a schematic flow chart showing the operation of neural network and Gaussian process regression on the elevation map and features included predictions using GPR models. In-situ measurements are made during the build and fed into the ensemble of GPR models. The predictions made by each GPR model are then combined to achieve a greyscale value to form an AI-generated porosity predictor.

[0076] FIG. 19 is a CT scan for ground truth validation of the printed sample, including example training data collection for Gaussian process regression and neural network training; features include low density areas, potential pore locations.

[0077] FIG. 20 is a CT scan for ground truth validation of the printed sample, including example training data collection for Gaussian process regression and neural network training; features include low density areas, potential pore locations.

[0078] FIG. 21 shows CT scans for ground truth validation of printed samples, including example training data collection for Gaussian process regression and neural network training; features include low density areas, potential pore locations, and low density areas identified with AI / ML.

[0079] FIG. 22 is a CT scan ground truth validation of the printed sample with example training data collection for Gaussian process regression and neural network training, features include low density areas, potential pore locations, and low density areas identified with AI / ML.

[0080] FIG. 23A is a graph showing printing anomalies observed using advanced mapping (difference between powder and melt) of layer 567, and FIG. 23B is a graph showing in-situ data and AI / ML and Gaussian process regression prediction testing of low density areas, low density locations and / or pore locations identified with AI / ML of slice 567, features include low density areas, potential pore locations, low density areas identified with AI / ML.

[0081] FIG. 24A is a graph of in situ data of low density locations and / or pore locations showing printing anomalies observed using advanced mapping (difference between powder and melt) of layer 567, and FIG. 24B is a graph of AI / ML and Gaussian process regression prediction tests of low density locations and / or pore locations showing low density areas identified with AI / ML of slice 567, where features include low density areas, potential pore locations, and low density areas identified with AI / ML.

[0082] FIG. 25 is a system diagram of a surveillance system including hardware / software elements.

[0083] FIG. 26 is a photograph of the monitoring system attached to a printer (laser powder bed fusion) showing the system configuration.

[0084] FIG. 27 is a photograph showing a streak pattern on a powder bed, features include powder areas, streak patterns.

[0085] FIG. 28 is a photograph showing the streak pattern on the powder bed (2), features include molten areas, powder areas, and streak patterns.

[0086] Figure 29A is a photograph showing the camera system. The camera is attached to an EOS M 290 powder bed fusion 3D printer. Figure 29B is a photograph showing the projector system. The projector is attached to an EOS M 290 powder bed fusion 3D printer.

[0087] Figure 30 is a photograph showing the camera system. The camera is attached to an exone innovent binder jetting 3D printer. The projector is attached to an exone innovent binder jetting 3D printer.

[0088] system As used in this application, the terms "component" and "system" are intended to refer to a computer-related entity, either hardware, a combination of hardware and software, software, or software in execution. For example, a component may be, but is not limited to, a process running on a processor, a processor, an object, an executable, a thread of execution, a program, and / or a computer. As an example, both an application running on a server and the server may be a component. One or more components may reside within a process and / or thread of execution, and a component may be localized on one computer and / or distributed between two or more computers.

[0089] Generally, program modules include routines, programs, components, data structures, etc. that perform particular tasks or implement particular abstract data types. Moreover, those skilled in the art will appreciate that the methods of the present invention may be practiced with other computer system configurations, including single-processor or multi-processor computer systems, minicomputers, mainframe computers, and personal computers, handheld computing devices, microprocessor-based or programmable consumer electronics, and the like, each of which may be operatively coupled to one or more associated devices.

[0090] The illustrated aspects of the invention may also be practiced in distributed computing environments where certain tasks are performed by remote processing devices linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.

[0091] A computer typically includes a variety of computer-readable media. Computer-readable media may be any available medium that can be accessed by a computer, including both volatile and nonvolatile media, removable and non-removable media. By way of example and without limitation, computer-readable media may comprise computer storage media and communication media. Computer storage media includes volatile and nonvolatile media, removable and non-removable media implemented in any method or technology for storage of information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory, or other memory technology, CD-ROM, digital versatile disk (DVD), or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage, or other magnetic storage devices, or any other medium that can be used to store the desired information and that can be accessed by a computer.

[0092] Communication media typically embodied computer-readable instructions, data structures, program modules, or other data in a modulated data signal, such as a carrier wave or other transport mechanism, and includes any information delivery media. The term "modulated data signal" means a signal that has one or more of its characteristics set or changed in such a manner as to encode information in the signal. By way of example, and not limitation, communication media include wired media, such as a wired network or direct-wired connection, and wireless media, such as acoustic, RF, infrared and other wireless media. Combinations of any of the above should also be included within the scope of computer-readable media.

[0093] Software includes applications and algorithms. Software may be implemented on a smartphone, tablet, or personal computer, in the cloud, on a wearable device, or on other computing or processing devices. Software may include logs, journals, tables, games, records, communications, SMS messages, websites, charts, interactive tools, social networks, VOIP (Voice Over Internet Protocol), email, and video.

[0094] In some embodiments, some or all of the functions or processes described herein are executed by a computer program formed from computer readable program code and embodied in a computer readable medium. The phrase "computer readable program code" includes any type of computer code, including source code, object code, executable code, firmware, software, etc. The phrase "computer readable medium" includes any type of medium that can be accessed by a computer, such as read only memory (ROM), random access memory (RAM), hard disk drive, compact disc (CD), digital video disc (DVD), or any other type of memory.

[0095] While the invention has been described in connection with various embodiments, it will be understood that the invention is capable of further modifications. This application is intended to cover any variations, uses, or adaptations of the invention that generally follow the principles of the invention, including departures from the present disclosure as come within known practice in the art to which the invention pertains.

Claims

1. 1. A method for detecting print defects in additive manufacturing, comprising: a. capturing an image of at least a portion of a surface of an object with a camera and a projection system; b. Obtaining a surface of said object in real time by projecting a pattern with an elevation map; c. using the resulting elevation map to detect printing defects in the powder layer or the newly solidified layer of said surface; A method comprising:

2. The method of claim 1 , further comprising evaluating the solidified layer based on melting, sintering, or bond solidification.

3. 10. The method of claim 1, further comprising using an elevation difference between the newly solidified layer and a previous powder layer to find print defects.

4. The method of claim 1 , further comprising: finding print defects using an elevation difference between a solid layer and a subsequent powder layer.

5. 10. The method of claim 1, further comprising using the measured print defects to find defects in the part through the use of linear regression, signal processing, anomaly counting artificial intelligence, finite element modeling, convolutional neural networks, or others.

6. The method of claim 5 , wherein the part defect comprises a density defect, a porosity defect, a metallurgical defect, a fusion bond failure, a crack, a deformation, or a dimensional defect.

7. 2. The method of claim 1, wherein the print defects include warping, over melting, under melting, thermal distortion, poor melt bonding, under powder feeding, uneven powder coating, recoater blade hopping, powder streaks, powder dimples, part extrusion, recoater blade defects, or build plate skew, recoater cross section information, build surface skew, plate skew, build plate flatness, build plate vibration, build plate vertical accuracy, powder coating thickness, powder coating roughness, powder coating streaks, powder coating hopping, powder coating under feeding, powder coating clumps, powder coating agglomerates, powder coating changes after melting or solidifying or bonding, uneven powder particles, thermal distortion, component extrusion, absolute layer height from nominal, material collapse, support structure distortion, total distortion, surface distortion, particle spitting, particle agglomeration, particle agglomeration, uneven melting, uneven sintering, laser spacing effect, laser spot size, laser power, binder penetration effect, binder powder interaction, binder deposit high solidification, or any other physical phenomenon.

8. 1. A method for detecting printing defects in an additive manufacturing process, comprising: a. capturing an image of at least a portion of a surface of an object with a camera and a projection system; b. using an image of at least a portion of a surface of the object in conjunction with an illumination system to encode spatial values ​​for each portion of the image during the additive manufacturing process; c. Using said spatial values ​​to estimate the uniformity of the encoded spatial phase of the powder layer or the freshly solidified layer and to locate print defects; A method comprising:

9. The method of claim 8 , further comprising evaluating the solidified layer based on melting, sintering, or bond solidification.

10. 10. The method of claim 8, further comprising using a spatial phase difference between the newly solidified layer and a previous powder layer to find print defects.

11. 10. The method of claim 8, further comprising finding print defects using a spatial phase difference between a solid layer and a subsequent powder layer.

12. 10. The method of claim 8, further comprising using the measured print defects to find defects in the part through the use of linear regression, signal processing, anomaly counting artificial intelligence, finite element modeling, convolutional neural networks, or others.

13. The method of claim 12 , wherein the part defect comprises a density defect, a porosity defect, a metallurgical defect, a fusion bond failure, a crack, a deformation, or a dimensional defect.

14. 9. The method of claim 8, wherein the print defects include warping, over melting, under melting, thermal distortion, poor melt bonding, under powder feeding, uneven powder coating, recoater blade hopping, powder streaks, powder dimples, part extrusion, recoater blade defects, or build plate skew, recoater cross section information, build surface skew, plate skew, build plate flatness, build plate vibration, build plate vertical accuracy, powder coating thickness, powder coating roughness, powder coating streaks, powder coating hopping, powder coating under feeding, powder coating clumps, powder coating agglomerates, powder coating changes after melting or solidifying or bonding, uneven powder particles, thermal distortion, component extrusion, absolute layer height from nominal, material collapse, support structure distortion, total distortion, surface distortion, particle spitting, particle agglomeration, particle agglomeration, uneven melting, uneven sintering, laser spacing effect, laser spot size, laser power, binder penetration effect, binder powder interaction, binder deposit high solidification, or any other physical phenomenon.