Open Access
Add to BibSonomy
Abstract
As in welding, directed energy deposition (DED) additive manufacturing (AM) generates complex residual stresses and distortions commensurate with the complexity of the scan pattern used for deposition. To date, measuring DED distortions on complex geometries has only been achieved post process, discarding the complex thermomechanical history that leads to that final material state. In this work, surround stereo digital image correlation (DIC) is used to 3D map surfaces and strain tensors in-situ in a powder-blown laser DED system. Infrared thermography is then projected onto these surfaces to record the full thermomechanical history of printed parts. DIC presents a unique challenge to DED AM, as no part exists at the beginning of deposition, which (a) prevents application of an appropriate speckle pattern and (b) denies the user a zero strain reference frame. Solutions to these problems are proposed and their limitations explored herein. In sum, this work presents a relatively low-cost solution to monitoring and optimizing the unique temporal artifacts induced by complex scan strategies that was previously unobtainable.
1. Introduction
Directed energy deposition (DED) is a form of additive manufacturing (AM) that permits rapid fabrication of metallic components that can meet wrought strength [1] while permitting unprecedented geometric design flexibility [2]. DED is a multi-pass welding process that generates complicated, nonuniform heating and cooling cycles, depending on the choice of scanning strategy. Thermal transfer in DED has been shown to be highly dependent on geometry and has many important consequences on process stability [3], microstructure [4], and mechanical properties [5]. One particular concern is the introduction of residual stresses and distortion from thermal expansion [6], which in many material systems induces cracking [7], substantial work, and dislocation generation [8], and a general inability to meet dimensional tolerances without post machining.
The severe consequences of this thermal nonuniformity have motivated a broad field of monitoring and control research to improve the capabilities of the process [9]. These capabilities can be divided into four categories: (1) point measurements such as thermocouples or strain gauges, (2) integrated measurements such as acoustic or photonic emission monitoring [10] [11], (3) local melt pool measurements such as coaxial thermal and multispectral imaging [12] [13], and (4) global measurements such as a thermal camera viewing the entire part. While the first three capabilities have been put to many successful uses, such techniques do not have the combination of spatiotemporal identity necessary to provide the history and long-term interaction of individual geometrical features with repeated thermal cycling.
In contrast, global off-axis imaging provides direct information about the history of various geometrical components. However, such measurements are hindered by the complexity of the component, as various 2D pixels must be related back to particular coordinates in the 3D CAD geometry. This frequently limits academic studies to simple geometries such as flat walls or coupons, for which manual registration of data is possible. In metal AM, this approach impedes the intended purpose of understanding the interactions between complex geometry, scan strategy, and thermomechanical history.
It becomes clear from the literature that a distinct classification can be made between available monitoring options. Geometrically arbitrary measurements must meet the criteria of spatially mapping measurements into the coordinate system of the printed object for objects of any shape. This distinction is important in the field of AM, because it defines a usability threshold where the technique can measure each and every part instead of parts being designed to fit the measurement technique.
In practice, geometrically arbitrary measurements in AM have been achieved in several ways, as described in the literature. As an example, structured light scanning techniques can provide depth information across the image, which permits any contained information in the image to be mapped spatially against the intended CAD geometry [14]. Holzmond et al. developed a top-down stereo DIC system that used polymer composite for contrast, which is capable of mapping identified defects to their spatial location in the component [15].
Focusing on DIC in metal DED, several prior studies have investigated DIC in-situ with AM, as shown in Table 1. Biegler et al. used DED to manufacture a 316L stainless steel thin-wall geometry and then spray painted a speckle pattern and printed ten additional layers while recording stereo video at 10 fps for full field DIC of a single surface [16]. Results from DIC and thermocouples compare reasonably well with thermomechanical AM finite element simulation, demonstrating DIC’s ability to provide validation data for models. However, this approach requires a process pause for spray painting, which allows the component time to cool, disrupting the process and limiting its usefulness in measuring the natural behavior of large and complex geometries.
Table 1. Comparison of prior and current work on DIC in DED AM.
Xie et al. similarly manufactured a thin wall of Ti-6Al-4V with DED, but instead they used natural surface roughness for the speckle pattern, using a fixed laser illumination source [17]. Noise levels using surface roughness as speckle are reported to be ±0.04%, and they demonstrate the ability to detect the compressive-to-tensile wave generated from the passing of the laser. However, this study was limited, as it used a single camera for 2D strain measurement only, limiting the geometries to which it can be applied. Whether using natural surface roughness of such a reflective material would be effective under stereo imaging remains unresolved.
In this study, in-situ DIC in AM was implemented in a number of ways to determine its usefulness for industrial monitoring of strain and temperature gradients. First, the optical requirements and limitations of using natural surface roughness in stereo were addressed, allowing for generation of 3D strain mapping of arbitrary shapes without prior patterning. The principles of this optical design motivated installation of multiple sets of stereo systems at relatively low stereo angles. This was implemented in-situ for a full 360-degree mapping of every part printed on the machine. Second, these stereo pairs of cameras were combined with low-cost long wave infrared (LWIR) microbolometer cameras for spatially resolved thermal measurement projected onto the 3D DIC strain map. Third, in AM, there is a unique problem for DIC in that there is no zero strain state for a reference image. Therefore, a solution is proposed herein using the end frame of the video as a reference and then inverting strains. Advantages and limitations of this approach are discussed in detail below. In summary, the demonstrated technique provides a path forward for obtaining previously unobtainable stress and thermal information in a geometrically arbitrary manner. This allows for data collection on every DED manufactured part which can be leveraged for myriad process optimization and control applications.
2. Experimental methods
2.1 Optical benchtop trials
As a preliminary validation step, stereo correlation quality on as-printed surfaces was measured in an optical benchtop experiment. A representative thin wall geometry was imaged under a variety of controlled view angles, camera stereo angles, and illumination angles as described in Fig. 1. Angles are referenced according to spherical notation from the axis established by (θ, φ) subscripted for the second camera (C), sample (S), and light (L). Cameras were FLIR Grasshopper 3 machine vision cameras with a 12.3 MP resolution and Rodagon 60mm prime lenses. Images were analyzed with commercial DIC code from Correlated Solutions (USA). The camera lens and stereo angle parameters were calibrated in accordance to DIC standard methods [18] using a 15 mm pitch dot grid plate.
Fig. 1. Optical table setup for investigating stereo and illumination angle effects on correlation accuracy, capable of independently varying rotation and tilt angles for the sample, second camera, and illumination.
2.2 Deposition machine setup
For all demonstration parts in this study, deposition was performed with a BeAM Modulo 400 machine (Strasbourg, FRA), using the 10VX nozzle design. Deposition parameters for trial geometries were maintained constant at the levels specified in the following. Power was set to 400 W and was calibrated with a PRIMES Cube L (GmbH, DEU). The laser power profile was a 610 µm diameter flat-top at the beam waist with a divergence of 0.78 degrees which was set to the nominal working distance of 3.5 mm as calibrated with an Ophir BeamCheck system. The scan speed was set to 2250 mm/min. The powder feed rate was maintained at 12 ± 1 g/min, and feedstock material was gas atomized stainless steel (SS) 316L with a size range of 46–105 µm, except for one functionally graded ring structure as presented in Fig. 6(c). This structure transitioned from a ferritic ANA-2 powder at the base for fifteen layers, followed by 50–50% ANA-2 and SS316L for 15 layers, followed by fifteen layers of SS 316L. For parts with infill hatch patterns, the pattern was rotated 57° between layers.
2.3 In-situ system instrumentation methods
Upon analysis of optical benchtop experiments, an in-situ system was designed and installed in a BeAM Modulo 400 machine as shown in Fig. 2. The machine was outfitted with four sets of cameras, and illumination was provided in the four corners of the build chamber, which are sufficiently compact so as to not interfere with operation of the machine. Each of the four sensor systems consisted of (a) two 20 MP rolling shutter visible cameras capable of capturing frames at 18 Hz and spectrally filtered with a broad bandpass filter to a 440–555 nm wavelength region with fixed focal length lenses, (b) a FLIR Boson microbolometer long-wave infrared camera with a resolution of 640×512 pixels, (c) a 320 J Xenon strobe, and (d) a 60 W white light-emitting diode (LED). The visible cameras and the strobe were capable of synchronous or asynchronous hardware triggering through a microcontroller. Systems were chosen and arranged to provide full coverage over a build volume of 400 mm diameter at 400 mm tall, with a projected pixel size of less than 100 µm to ensure resolution of specular highlights from individual unmelted, adhered powder particles.
Fig. 2. In-situ DIC + IR setup. (a) Five axis build chamber with four camera sets installed in corners. (b) camera set with two 20 MP visible light cameras, one Long Wave Infrared microbolometer camera, Xenon strobe light and 60W LED. (c) Single sample image from a visible camera under LED illumination showing particle and layer contrast, and (d) sample LWIR image showing thermal dissipation from the most recently printed layer.
After camera installation, camera lens parameters and stereo transformations were calibrated for all visible and infrared cameras using standard DIC methods with a heated aluminum dot grid pattern with 15 mm pitch [18].
The temperature-intensity response for the FLIR Bosons with 14 mm and 18 mm lenses was calibrated using a blackbody furnace. Emissivity values for rough and AM SS316L surfaces were drawn from the literature [19] and were assumed to be constant at 0.3. The accuracy and potential improvements of these values and assumptions is discussed in the analysis.
3. Results
3.1 Benchtop investigation of the stereo correlation of natural surface roughness
In designing an in-situ system, there are tradeoffs between the attainable field of view, the resolution of the image and sensor, the data acquisition frequency, the cost of the system, and the cost of processing data acquired from the system. As such, it is of high practical importance to understand the sensitivity of a system design on naturally occurring surface roughness.
For reflective surfaces, contrast is generated primarily through differences in specular reflection due to variation in the incident angle between the illumination source, the target surface, and the camera. To characterize the surface roughness typical of DED, as presented in Fig. 3, a high magnification confocal image shows that two primary surface features generate illumination variation: incompletely melted, welded-on powder particles, as well as the layer-to-layer “scalloping” caused by the surface tension of the melt pool. These roughness features vary spatially in the process, even when controlled for all nominal process parameters, as evidenced by the different oxide coloration on the part and the distribution of the roughness features.
Fig. 3. Confocal surface topology measurement of sample 316L cylinder. Two sources of roughness are observed: partially melted powder particles and layer-to-layer “scalloping” from curvature induced by melt pool surface tension. Different regions of the part demonstrate varying degrees of each form of roughness, despite constant nominal process conditions, due to heat accumulation.
Each of these roughness features will generate a correlatable roughness pattern under sufficient magnification and illumination conditions; however, the correlation quality may depend on these factors. To simulate in-situ DIC using these surface features, the high-resolution images of a cylindrical specimen were binned down to larger synthetic pixels (2×2 to 20×20 in single pixel steps) and were processed with DIC, as shown in Fig. 4.
Fig. 4. Sample synthetically de-magnified images and correlation results from optical benchtop trials; (a) one camera view at the original pixel magnification (17.9 µm) which was then binned from 2×2 to 20×20; 3×3 binning in (b) 3D and (c) mapped 2D correlation quality (sigma) results.
At high resolutions, individual powder particles and layer lines are clearly visible, but they become obscured with decreasing pixel magnification. Four regions in Fig. 4 were chosen for comparison across magnifications, corresponding to the two different dominant roughness regimes (scalloping vs. particle) and illumination condition (brightfield vs. darkfield). The effect of magnification on correlation quality (sigma) is plotted in Fig. 5. Sigma is a correlation quality metric that estimates measurement inaccuracy from residual mismatch between reference and deformed images. Correlation parameters were set to a 125 pixel subset size sampling every four pixels, with no threshold for correlation quality.
Fig. 5. Synthetic magnification effect on correlation quality of natural surface roughness for regions highlighted in Fig. 4 with brightfield and darkfield illumination and roughness regimes of frequent attached powder particles and large layer scalloping.
As can be seen in Fig. 5, for all regions, a minimum correlation quality is achieved not actually at the highest magnification (17.9µm), but at a 3×3 pixel binning (53.7 µm), after which correlation quality steadily degrades. Degradation of quality for high binning is expected, as less information is present in a given window to localize and lock in the correlation, but counterintuitively, the highest resolution has a worse score. This could possibly be attributed to sensor noise which would be somewhat smoothed through binning, or it could be due to the differential shift in specular reflection off of individual roughness features. These possible causes and their implications are discussed further in the analysis section, but in practical terms, this cues designing the imaging system’s resolution to match but not exceed the resolution required to resolve individual powder particles (45–105µm). Furthermore, it highlights that higher correlation quality can be achieved under darkfield illumination; however, successful reconstruction is at least possible under a variety of surface/illumination incident angles as long as the camera stereo angle is low, allowing relative design freedom in illumination placement.
3.2 In-situ DIC from surface roughness
After benchtop optical testing was complete, the system shown in Fig. 2 was designed and installed. Multiple parts of varying geometries were printed and recorded with the DIC + IR system, as shown in Fig. 6.
A demonstration prismatic containment structure made of SS316L and 220 mm in diameter was printed and recorded for analysis. An image from one camera system with overlaid DIC results is shown in Fig. 7. Two columns of inspection points are shown in Fig. 7(a), row A, following the center of a thin double-bead thick wall, and column B, following a wall supported from internal latticework. The deformations of these sets of points are plotted over time in Fig. 7(d).
Fig. 6. 3D topography from in-situ stereo correlation of natural surface roughness for (a) a 220 mm diameter demonstration containment structure, (b) a 60×30 mm oblong block, (c) a 50 mm dia. 5 mm thick wall ring, and (d) “vase” geometry with a single track spiral printed with a 5 axis toolpath rotating and tilting the substrate to avoid overhangs.
Fig. 7. (a) In-situ DIC measurement from a single perspective using the final frame of the video as a strain reference and mapping total XYZ displacement from the reference of the final frame of the recording. Inspection points are overlaid. (b) 3D mapped infrared intensity overlaid on a visible image showing heat concentration from prior deposition at the large interior pockets. (c) Correlation quality map showing an error estimate ranging up to 0.034 pixels (12.1 µm equivalent in Z). (d) Inspection points showing total displacement over 40 hours of printing. Two breaks in the recording for data transfer are labeled with black dotted lines, and three pauses in printing for powder hopper refills are labeled with red arrows.
3.3 3D mapped thermal measurement
To obtain the full thermomechanical history of printed parts, the emitted light signal from the four longwave infrared cameras was mapped to the 3D geometry produced from DIC for a variety of geometries. IR images recorded during printing of the demonstration containment structure in Fig. 6(a) are shown in Fig. 8. A more quantitative thermal recording was performed on the functionally graded ring structure shown in Fig. 6(c), as shown in Fig. 9; absolute temperature and part emissivity was calibrated through comparison with a type K thermocouple tack welded to the build substrate.
Fig. 8. (a) Sample longwave infrared intensity image frame. (b) Mapping of IR intensity spatially onto 3D DIC results; calibration of visible and IR camera intrinsic and extrinsic parameters allows for mapping IR intensity spatially onto 3D DIC results for any given image frame.
Fig. 9. (a) Infrared thermal measurements from the 50 mm ring geometry presented in Fig. 6(c). (b) Comparison of temperature history from four points on the part sidewall. (c) Two peaks per layer are observed for contours and infill.
4. Analysis
4.1 DIC speckle distortion from specular reflections
One underlying assumption of DIC is that the target’s unique speckle pattern reflects light diffusely, allowing for the pattern to be identified from multiple viewing and illumination angles through homographic transformations. This is violated to a degree when using natural surface roughness of DED AM components, as contrast is generated primarily through variations in specular reflection off of surface features. As the camera and illumination angle changes, the apparent position of the specular highlight on rounded features shifts. For spherical and cylindrical samples, this can be expressed with a simple geometric function according to Eq. (1) and Fig. 10:
Where ${d_{shift}}$ is the apparent shift in particle position, r is the radius of the shiny roughness feature, and ${\theta _C}$ and ${\theta _L}$ are the differences in camera view angle and illumination angle from the reference camera, similar to the setup in Fig. 1. Equation (1) can then be plotted over a variety of camera and illumination angles, as shown in Fig. 11.
Fig. 10. Reflections off of spherical particles will change their apparent position with changing camera and illumination angles according to Eq. (1) The magnitude of this shift is a simple geometric function. (c) The sensitivity of this function is plotted in Fig. 11 for a sample 80 µm diameter particle, illustrating the motivation for using low stereo angles for correlation.
Fig. 11. Apparent shift in speckle pattern location plotted as a function of camera and illumination angle, per Eq. (1).
From this equation, several design principles to minimize distortion due to speckle shift can be expressed. First, correlation accuracy will vary directly with the size of roughness features observed, and correlation quality will vary directly with the dispersion of feature size. This is confirmed by observing that in Fig. 4 and Fig. 5, correlation quality degrades in areas where contrast is primarily driven by layer to layer ‘scalloping’ (∼200 µm) as opposed to regions where contrast is driven by powder particles (∼70 µm). Second, error from specular speckle shift does not depend strongly on the angle of illumination; therefore illumination angles can be chosen freely to highlight the smallest, highest contrast surface features. Third, correlation error from speckle shift depends strongly on camera stereo angle. This, combined with the possibility of the intensity of reflections off of powder particles changing intensity (responsible for the failures in correlation and data holes in Fig. 6(a) and (d)), drives design to use very low camera stereo angles. This competes directly with conventional DIC practice of using higher (30–45°) stereo angles, which helps to mitigate depth measurement error [18]. In practice, the system achieves Z-depth correlation quality of 10–25 µm (as opposed to 3–10 µm for X and Y), which for this application is an acceptable tradeoff.
4.2 Temperature accuracy in DED AM with microbolometer cameras
Due to the nonlinear relationship between temperature and recorded infrared intensity, it is required to both carefully calibrate the sensor’s intensity response function and measure target object emissivity. The spectrum of emitted blackbody radiation follows Planck’s law, modified for a graybody:
Fig. 12. Measured intensity from a blackbody furnace for two FLIR Boson 640 microbolometers, with two different lenses, compared to the minimum ambient background intensity read from the DED chamber. Error bars represent pixel to pixel variance within the same frame.
Object emissivity was measured by comparing intensity readings during printing to a thermocouple attached to the substrate, as shown in Fig. 13. Good agreement between the thermocouple and the IR camera was found when the substrate’s emissivity was set to 0.19; if the substrate were assumed to be uniformly emissive and the deposited material directly adjacent to the substrate were assumed to be the same temperature, then the measured emissivity of printed surfaces near the substrate is found to be ∼0.23–0.25. This is well within the bounds of what could be expected from lightly oxidized and rough SS316L surfaces [19].
Fig. 13. Comparison of in-situ microbolometer infrared camera temperature readings vs. thermocouple readings. (a) Variability in fit quality due to occlusion of original substrate surface with rougher, more emissive loose powder (inset visible images). (b) Comparison of two readings showing distinct bands for heating and cooling.
These assumptions allow for a presentable estimation of target temperature, but it is important to note that emissivity is a function of temperature, surface roughness, oxidation, and material [19], all of which may vary according to machine parameters and conditions, as demonstrated in Fig. 3. As such, any spatial or temporal variance in the target’s emissivity will directly degrade accuracy of the IR thermal measurement. For instance, as shown in the inset images of Fig. 13(a), during deposition, the substrate became periodically covered with loose powder particles, which present as a rougher, more emissive surface, introducing inaccuracy into even the calibration measurement. As processing conditions are typically unique for every build, an automated, spatially sensitive method for measuring emissivity is clearly desirable. In future work, multi-band techniques may be applied such as those presented by Dagel [21] to spatially map emissivity and to provide improved temperature measurement accuracy.
4.3 DIC reference frame problem in AM
Unlike in traditional DIC measurements, a fundamental problem arises from attempting DIC in-situ on any AM process—namely, that there is no part at the beginning of the process, so it is impossible to obtain a zero strain reference frame. Prior attempts use an arbitrary time part way into the build process as a reference either by fully stopping the process and spray painting a speckle pattern onto the part [16] or by pausing the build and capturing a reference frame [17]. As illustrated in Fig. 14(c), the issue with these approaches is that it is impossible to consider any of the material added after the reference frame. These approaches only measure the strain induced on layers visible in the reference frame. Because the material has already undergone significant deformation, it is not possible to retrieve the full strain history of the material.
Fig. 14. Comparison of reference frame options for in-situ DIC for AM. (a) Schematic illustrating inner and outer surfaces visible in (b) real images. (c) Comparison of others’ work using an arbitrary reference frame—which cannot extract information for new material deposited—to the proposed methods of using the final frame as a reference (d).
To address this systemic methodological gap, an alternate approach was explored in which the final image of the part after the process has finished is used as the reference for the entire time series, as illustrated in Fig. 14(d). Therefore, all deformations measured are relative to the cold final state, and generated plots must be interpreted considering this assumption. As shown in Fig. 7(d), using this reference frame allows for tracking the full-field deformation, but the part will appear to start in a highly deformed state and will converge toward the final cold state over time. Therefore, analysis of the total deformation of each identified point requires the reference position of the data to be updated and inverted to refer to the first frame at which material was present.
This approach provides direct information about relative part deformation, but it falls somewhat short in that it only allows for consideration of the part’s outer surfaces which are visible throughout the build process. Furthermore, the final image of a frame is not available for analysis until the end of the build, which prevents this technique from being used in real time. In future work, it may be possible to automate reference frame initialization on a layer-by-layer basis to allow real time analysis and improved part coverage.
5. Conclusions
An in-situ digital image correlation system for measuring thermomechanical history in directed energy deposition additive manufacturing was developed. The system leverages naturally occurring surface roughness as a correlation pattern, providing a minimally invasive yet high resolution evaluation of deposition performance. Infrared imaging was projected onto the generated 3D map, allowing for direct estimation of part temperature history. Generated information from this relatively low-cost system offers an unprecedented window into the dynamic interactions between the complicated scan strategies used for deposition, thermal absorption, dissipation, and the material’s resultant expansion and contraction. Ultimately, the method has implications for directly measuring thermomechanical properties and microstructure-relevant information and mapping it spatially to the deposited component and is expected to yield a rich dataset for future modeling and process control efforts.
Funding
Office of Nuclear Energy; Fossil Energy, Inc; Oak Ridge National Laboratory (LOIS 9282).
Acknowledgments
This research and imaging system was sponsored by the Transformational Challenge Reactor (TCR) program and supported by the US Department of Energy, Office of Nuclear Energy. The publication of this research was supported by the US Department of Energy, Advanced Manufacturing Office. Data was included from deposited components performed for research sponsored by Fossil Energy, Inc. The AM system was purchased with Laboratory Directed Research & Development funding (LOIS 9282), and is located at the U.S. Department of Energy’s Manufacturing Demonstration Facility, at Oak Ridge National Laboratory.
This manuscript has been authored by UT-Battelle, LLC, under contract DE-AC05-00OR22725 with the US Department of Energy (DOE). The US government retains and the publisher, by accepting the article for publication, acknowledges that the US government retains a nonexclusive, paid-up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript, or allow others to do so, for US government purposes. DOE will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan (http://energy.gov/downloads/doe-public-access-plan).
Disclosures
The authors declare no conflicts of interest.
References
1. J. J. Lewandowski and M. Seifi, “Metal additive manufacturing: a review of mechanical properties,” in Annual Review of Materials Research 46, D. R. Clarke, Ed. Palo Alto: Annual Reviews, 2016, pp. 151–186.
2. D. D. Gu, W. Meiners, K. Wissenbach, and R. Poprawe, “Laser additive manufacturing of metallic components: materials, processes and mechanisms,” Int. Mater. Rev. 57(3), 133–164 (2012). [CrossRef] .
3. J. C. Haley, B. Zheng, U. S. Bertoli, A. D. Dupuy, J. M. Schoenung, and E. J. Lavernia, “Working distance passive stability in laser directed energy deposition additive manufacturing,” Mater. Des. 161, 86–94 (2019). [CrossRef] .
4. Y. Lee, M. Nordin, S. S. Babu, and D. F. Farson, “Effect of fluid convection on dendrite arm spacing in laser deposition,” Metall. Mater. Trans. B 45(4), 1520–1529 (2014). [CrossRef] .
5. M. E. Stender, L. L. Beghini, J. D. Sugar, M. G. Veilleux, S. R. Subia, T. R. Smith, C. W. San Marchi, A. A. Brown, and D. J. Dagel, “A thermal-mechanical finite element workflow for directed energy deposition additive manufacturing process modeling,” Addit. Manuf. 21, 556–566 (2018). [CrossRef] .
6. P. Promoppatum and S.-C. Yao, “Influence of scanning length and energy input on residual stress reduction in metal additive manufacturing: numerical and experimental studies,” J. Manuf. Process. 49, 247–259 (2020). [CrossRef] .
7. T. Mukherjee, W. Zhang, and T. DebRoy, “An improved prediction of residual stresses and distortion in additive manufacturing,” Comput. Mater. Sci. 126, 360–372 (2017). [CrossRef] .
8. T. R. Smith, J. D. Sugar, C. San Marchi, and J. M. Schoenung, “Strengthening mechanisms in directed energy deposited austenitic stainless steel,” Acta Mater. 164, 728–740 (2019). [CrossRef] .
9. S. K. Everton, M. Hirsch, P. Stravroulakis, R. K. Leach, and A. T. Clare, “Review of in-situ process monitoring and in-situ metrology for metal additive manufacturing,” Mater. Des. 95, 431–445 (2016). [CrossRef] .
10. S. A. Shevchik, G. Masinelli, C. Kenel, C. Leinenbach, and K. Wasmer, “Deep learning for in situ and real-time quality monitoring in additive manufacturing using acoustic emission,” IEEE Trans. Ind. Inf. 15(9), 5194–5203 (2019). [CrossRef] .
11. M. Montazeri, A. R. Nassar, A. J. Dunbar, and P. Rao, “In-process monitoring of porosity in additive manufacturing using optical emission spectroscopy,” IISE Trans. 52(5), 500–515 (2020). [CrossRef] .
12. S. Ocylok, E. Alexeev, S. Mann, A. Weisheit, K. Wissenbach, and I. Kelbassa, “Correlations of melt pool geometry and process parameters during laser metal deposition by coaxial process monitoring,” in 8th International Conference on Laser Assisted Net Shape Engineering (lane 2014)56, pp. 228–238, M. Schmidt, F. Vollertsen, and M. Merklein, Eds. Amsterdam: Elsevier Science (2014).
13. S. Donadello, M. Motta, A. G. Demir, and B. Previtali, “Coaxial laser triangulation for height monitoring in laser metal deposition,” Procedia CIRP 74, 144–148 (2018). [CrossRef] .
14. I. Garmendia, J. Leunda, J. Pujana, and A. Lamikiz, “In-process height control during laser metal deposition based on structured light 3D scanning,” Procedia CIRP 68, 375–380 (2018). [CrossRef] .
15. O. Holzmond and X. Li, “In situ real time defect detection of 3D printed parts,” Addit. Manuf. 17, 135–142 (2017). [CrossRef] .
16. M. Biegler, B. Graf, and M. Rethmeier, “In-situ distortions in LMD additive manufacturing walls can be measured with digital image correlation and predicted using numerical simulations,” Addit. Manuf. 20, 101–110 (2018). [CrossRef] .
17. R. Xie, “The full-field strain distribution and the evolution behavior during additive manufacturing through in-situ observation,” Mater. Des. 150, 49–54 (2018). [CrossRef] .
18. M. A. Sutton, J. J. Orteu, and H. Schreier, Image Correlation for Shape, Motion and Deformation Measurements: Basic Concepts,Theory and Applications. Springer US (2009).
19. T. S. Hunnewell, “Total hemispherical emissivity of SS 316L with simulated very high temperature reactor surface conditions,” Nucl. Technol. 198(3), 293–305 (2017). [CrossRef] .
20. C. Honniball, “Spectral response of microbolometers for hyperspectral imaging,” Masters Thesis, University of Hawai’i at Manoa, (2017).
21. D. J. Dagel, G. D. Grossetete, D. O. MacCallum, and S. P. Korey, “Four-color imaging pyrometer for mapping temperatures of laser-based metal processes,” in Thermosense: Thermal Infrared Applications9861, 986103 XXXVIII (2016).
