e-dokumen.com

attain a higher final density. The printing process has a build rate of 166 cm3/h. The manufacturingprocess is done in five steps: (1) each layer is configured with a specific set of process parameters; (2) arolling mechanism covers the build plate with a layer of powder; (3) the binder deposition processoperates similarly to the ink jet printing process: the print head moves in a pre-defined pattern over thebuild plate, depositing binder only when needed; (4) a radiative heating element scans the build plate todry the binder; (5) once the layer completes, the stage supporting the build plate lowers to the desiredlayer thickness (typically 50 to 200 µm), and the process repeats until completion of the geometry. Bydefault, the Innovent system is only equipped with standard sensing technologies to ensure its correctmechanical operation. There is no standard image base in situ monitoring available to assess the layer ofpowder or the quality of the binder deposition.The Innovent system is nearly identical to the Innovent system. The primary difference between the twosystems are their powder depositions mechanisms. The Innovent uses an eccentricly weighted motor tovibrate the powder hopper while the Innovent uses an ultrasonic driver instead.Figure 1. ExOne – BinderJet Innovent.2.2Figure 2. ExOne – BinderJet M-Flex.EXONE – M-FLEXThe M-Flex system uses the same technology and manufacturing process as the Innovent system(Figure 2). However, the M-Flex has a build volume of 400 250 250 mm3 (about 37 times the volumeof the Innovent). The printing process is very fast with a build rate of 1600 cm3/h (10 times the build rateof the Innovent and 13 times the build rate of the X-Line). Like the Innovent, the M-Flex has limitedinstrumentation.3.3.1IN SITU PROCESS MONITORINGEXONE INNOVENT AND INNOVENT The ExOne Innovent and Innovent systems are Binder Jet printers. One ExOne Innovent and one ExOneInnovent are located on Main Campus in Building 4508. Their internal designations are“ExOneInnovent-0137” and “ExOneInnovent-0166,” respectively. These machines produce several insitu data streams which are collected and analyzed as part of the TCR program. The following sub2

sections describe each of these data streams, highlighting any changes in hardware, software, or analysismethodologies in FY 2020. The in situ data are analyzed and visualized, together, using the Peregrinesoftware package, which is described in the Digital Platform supplementary material.3.1.1Layer-wise Powder Bed ImagingA 20 mega-pixel (MP) 8-bit grayscale camera, sensitive in the visible spectrum, captures an image of theentire print area immediately after binder deposition and powder spreading for each layer. A top-mountedlight strip provides illumination. This system was installed and is maintained, by ORNL. Image capturesequencing is controlled by Peregrine. Examples of a post-binder image and a post-spreading image forthe same layer of a build are shown in Figure 3.Figure 3. An image taken after binder deposition (left) and an image taken after powder spreading (right).This imaging system has not changed since FY 2019 and has an effective resolving power ofapproximately 90 µm. Neural network analysis of these powder bed images, performed by Peregrine,enables the detection of several types of defects and process flaws [1] as well as comparisons between theas-printed geometry and the intended geometry. Detectable anomalies include contamination of therecoating roller, improper powder spreading, and dragging of the bound powder (due to over saturation).Ongoing, early-stage research (not supported by TCR) at the Manufacturing Demonstration Facility3

(MDF) is seeking to directly quantify binder saturation levels based on imaging data. If successful, thistechnique may provide enhanced capabilities for the TCR program in the future.For comparison to the intended geometry, computer-aided design (CAD) information is first convertedinto a standard triangle language (STL) format and then “sliced” into layer images by the Arcam BuildAssembler software. Development of an alternative ORNL-developed slicing software was started in FY2020 and will continue in FY 2021. Note that the accuracy of the geometrical comparisons is limited bythe resolving power of the camera as well as the registration error between the camera data and thegeometry data. Due to the mounting configuration of the camera, data co-registration is less accurate forthe ExOne Innovent printers compared with the ConceptLaser M2 printer.3.1.2Temporal Log File DataAt the end of each build, a log file is produced that reports various machine error states as well astemporal sensor streams, including powder bed surface temperature and various process settings whichmight be altered by the operator during the build. This system is installed and maintained by ExOne, andnew in FY 2020, these data can be viewed using Peregrine, as shown in Figure 4.Figure 4. Example log file data displayed as a time series.3.1.3Metadata TrackingBuild-specific metadata are tracked using a combination of Peregrine and, new in FY 2020, the DigitalTool web interface. Tracked metadata now include the build name, build start date, build end date,machine-specific processing parameters, data sensitivity levels (e.g., export control restrictions), projectand customer information, feedstock material type, feedstock batch information, printer operatorinformation, ambient environmental data, and various operator-input notes. In FY 2021, the DigitalTool’s capabilities will continue to be expanded, allowing for increased data search functionality and theability to view Timelines of feedstock batch utilization and printer calibration and maintenance historiesas they relate to each build. A standardized procedure for capturing high-quality post-build images of theprinted parts is a goal for FY 2021.4

3.1.4Data TransferCurrently, the log file data are manually recovered from the ExOne Innovent and Innovent using anexternal hard drive, following each build. This hard drive is then hand-delivered to one or more desktopcomputers located directly adjacent to the printers, and the data are uploaded to the Savitar Data StorageServer. Metadata are tracked using both Peregrine and the Digital Tool web interface. Powder bedimaging data are capture and analyzed in real-time by Peregrine. The raw imaging data and analysisresults are upload automatically to the Data Storage Server.3.1.5Remote MonitoringPeregrine allows machine operators to monitor the status of the ExOne Innovent prints remotely throughautomated email/text alerts as well as a portal (Figure 5) showing the current and previously printedlayers with any detected defects highlighted. This system was new at the end of FY 2019 and has beenused routinely throughout FY 2020.Figure 5. Remote monitoring of an ExOne Innovent build using Peregrine.3.1.6Data Visualization and AnalysisThe in situ data produced by the ExOne Innovent and Innovent printers are analyzed by trained machineoperators and data team members using the Peregrine software package. A total of 282 ExOne Innoventbuilds were analyzed in FY 2020. A broader overview of Peregrine is provided in the Digital Platformsupplementary material. To demonstrate the usage of Peregrine in the context of the ExOne Innovent asingle case study involving a single build of a prototype reactor fuel element, is discussed below.Peregrine uses a neural network to analyze and classify each pixel in every layer of powder bed imagingdata (Section 3.1.1). One way to visualize these results is by overlaying the classifications on top of thepowder bed images, as shown in Figure 6. These classifications can be cumulatively analyzed throughoutthe entirety of the build and visualized as “heat maps.” Figure 7 shows a heat map highlighting5

“streaking” caused by contamination of the recoating roller. For this particular machine, this type ofstreaking is commonly observed in this region of the powder bed. In this build the streaking most likelyinterfered with one of the fuel elements, designated P3.Figure 6. Anomaly detections overlaid on top of apost-binder image of the powder bed.Figure 7. Heat map (projection) of recoater streakingdetections throughout the height of the build.The recoater streaking can also be observed in a 3D reconstruction of the build data (Figure 8). Viewingthese data temporally (Figure 9), it can be seen that the build initially started out without any rollercontamination, but recoater streaking began to appear at layer 257. The operator attempted to correct thiserror on two occasions, indicated by the two system pauses; however, the recoater streaking continued tooccur. Also note that the powder bed temperature drops by approximately 7 C after each system pause.This could have implications for binder drying rates on subsequent layers.6

Figure 8. A 3D reconstruction of the entire build.Figure 9. Time series visualization of recoater streaking detections throughout the height of the build.3.2EXONE M-FLEXA single ExOne M-Flex Binder Jet printer is located at the MDF in the low bay. This machine isinternally designated “ExOneMFlex-0126” and produces several in situ data streams which are collectedand analyzed as part of the TCR program. The following sub-sections describe each of these data streams,7

highlighting any changes in hardware, software, or analysis methodologies in FY 2020. The in situ dataare analyzed and visualized, together, using the Peregrine software package, which is described in theDigital Platform supplementary material.3.2.1Layer-wise Powder Bed ImagingA 10 MP 8-bit grayscale camera, sensitive in the visible spectrum, captures an image of the entire printarea immediately after binder deposition and powder spreading for each layer. A top-mounted light stripprovides illumination. Additionally, a 0.3 MP 8-bit mid-wave infrared (MWIR) camera captures an imageof approximately 80% of the build area immediately after binder deposition and powder spreading foreach layer. This system was installed and is maintained by ORNL. Image capture sequencing is controlledby Peregrine.Due to ORNL environmental health and safety concerns, movement of the M-Flex printer to a newbuilding, COVID-19 restrictions, and TCR funding restrictions, this camera system has not beenreinstalled in CY 2020. Reinstallation of the system is a priority for the beginning of FY 2021. Duringreinstallation, the MWIR camera will be moved to ensure that the entire powder bed is observable, andthe system will be modified to protect the optics from the abrasive silicon carbide powder. Examples of apost-binder image and a post-spreading image from each camera for the same layer of a build are shownin Figure 10.Neural network analysis of these powder bed images, performed by Peregrine, enables the detection ofseveral types of defects and process flaws. Detectable anomalies include contamination of the recoatingroller, improper powder spreading, and dragging of the bound powder (due to over saturation). Ongoing,early-stage research (not supported by TCR) at the MDF is seeking to directly quantify binder saturationlevels based on imaging data. If successful, this technique may provide enhanced capabilities for the TCRprogram in the future. Comparison of the as-printed geometry to the intended geometry has not beendemonstrated for this printer but is expected to be possible upon reinstallation of the imaging system inFY 2021.8

Figure 10. A visible-light image taken after binder deposition (top-left), a visible-light image taken afterpowder spreading (top-right), a MWIR image taken after binder deposition (bottom-left), and a MWIRimage taken after powder spreading (bottom-right).9

3.2.2Temporal Log File DataAt the end of each build, a log file is produced that reports various machine error states as well astemporal sensor streams, including powder bed surface temperature and various process settings whichmight be altered by the operator during the build. This system is installed and maintained by ExOne, andnew in FY 2020, these data can be viewed using Peregrine, as shown in Figure 11.Figure 11. Example log file data displayed as a time series.3.2.3Metadata TrackingBuild-specific metadata are tracked using a combination of Peregrine and, new in FY 2020, the DigitalTool web interface. Tracked metadata now include the build name, build start date, build end date,various processing parameters, data sensitivity levels (e.g., export control restrictions), project andcustomer information, feedstock material type, feedstock batch information, printer operator information,ambient environmental data, and various operator-input notes. In FY 2021, the Digital Tool’s capabilitieswill continue to be expanded, allowing for increased data search functionality and the ability to viewTimelines of feedstock batch utilization and printer calibration and maintenance histories as they relate toeach build. A standardized procedure for capturing high-quality post-build images of the printed parts is agoal for FY 2021.3.2.4Data TransferCurrently, the log file data are manually recovered from the ExOne M-Flex using an external hard drive,following each build. This hard drive is then hand delivered to a desktop computer located directlyadjacent to the printer, and the data are uploaded to the Data Storage Server. Metadata are tracked usingboth Peregrine and the Digital Tool web interface. Powder bed imaging data are capture and analyzed inreal-time by Peregrine. The raw imaging data and analysis results are upload automatically to the DataStorage Server.10

3.2.5Remote Monitoring and Automated Process ControlPeregrine allows machine operators to monitor the status of the ExOne M-Flex prints remotely throughautomated email/text alerts as well as a portal (Figure 5) showing the current and previously printedlayers with any detected defects highlighted. Initial experiments using silicon carbide powder on theExOne M-Flex suggest that mid-build process parameter modifications will be essential in order to ensuresuccessful prints. Because these prints may be longer than a day in duration, it is not feasible for thetechnicians to monitor and control these prints in person.New in FY 2020, Peregrine allows operators to remotely modify the process parameters mid-build for theExOne M-Flex. Similarly, Peregrine is also capable of autonomously modifying the process parametersduring the build. These capabilities were first demonstrated in FY 2019, however, implementation hasstalled due to new network security protocols at ORNL. Continuing these remote and autonomous processcontrol efforts is a priority for FY 2021.3.2.6Data Visualization and AnalysisThe in situ data produced by the ExOne M-Flex are analyzed by trained machine operators and data teammembers using the Peregrine software package. A total of six ExOne M-Flex builds were analyzed inFY 2020. Because this machine was non-operational for the majority of CY 2020, only limited analysisresults are available for FY 2020; however, results and capabilities are expected to be highly similar tothose reported for the ExOne Innovent machine. A broader overview of Peregrine is provided in theDigital Platform supplementary material.3.3CHEMICAL VAPOR INFILTRATION SYSTEMSThere are currently two operational lab-scale Chemical Vapor Infiltration (CVI) systems at ORNL,internally designated “OakRidgeLabScaleCVI-4515L104” and “OakRidgeLabScaleCVI-4508265.” Thefollowing discussions regarding data collection, tracking, and analysis relate to these two systems. Thefirst small-scale industrial CVI system will come fully online

companion report will discuss monitoring of the laser powder bed fusion (LPBF) AM process also used by the TCR program. This document is structured as follows: the first section gives a brief overview of the two binder jet machines, explaining the manufacturing proces