Laser based directed energy deposition system for operando synchrotron x-ray experiments

Additive manufacturing (AM), also known as 3D printing, is a bottom-up technique that prints three-dimensional materials using computer-aided-design (CAD).¹ There are numerous techniques used in metal AM, such as selective laser melting (SLM), selective electron beam melting (SEBM), and directed energy deposition (DED).^2–9 This paper focuses on powder fed DED, which has garnered interest for the capability in printing metallic alloys and repairing parts.¹⁰ In DED systems, a feedstock feeding mechanism is attached to a deposition head while the concentrated high-power energy source (laser, electron beam, plasma, or electric arc) simultaneously melts the material, forming a melt pool on the substrate.^3,10,11 Powder feed with a laser beam is the frequent choice for printing medium sized parts with complex geometries. The process can also print multiple powders concurrently and has a relatively faster cooling rate, on the order of 10³–10⁵ °C/s.^11–13 

As a result of the highly non-equilibrium nature of the process, DED has inherent challenges found in metal AM, including cracking, porosity formation, residual stress generation, and non-equilibrium phase transformations during solidification.^3,14,15 Broadly speaking, the challenges of DED can be categorized into two major aspects: (i) complex heat and mass transfer and (ii) microstructure evolution during thermo-mechanical processing. Complex transport phenomena in DED lead to defects (i.e., keyholing, lack of fusion, and mid-porosity zone), in addition to compositional segregation and phase transformation.^1,6,14,16 To understand such a non-equilibrium thermal process leading to defects, intensive numerical models, such as the finite element method, are employed. However, such methods are computationally expensive and may fail to accurately account for the complex heat and mass transfer observed during AM.¹⁶ Operando characterization using synchrotron x-ray sources is emerging as a real-time monitoring technique to develop a fundamental understanding of the underlying physical phenomena in AM.^17,18 The progress in this field has significantly evolved from studies focused on laser heating of solid metallic alloys^17,19 to more recent studies that have successfully designed AM process replicators that mimic the true nature of the process. The pioneering works in this domain mainly focused on synchrotron x-ray imaging of SLM.^18,20–23 Some diffraction studies were also performed in either reflection geometries to understand the effects of changing processing parameters and scanning strategies²⁴ or in transmission geometry to understand the evolution of residual stresses during the printing process.²⁵ More recent studies on diffraction explore the effects of thermal gradients during SLM more extensively^26,27 with similar studies for DED also emerging, such as designed DED replicators to perform synchrotron imaging and diffraction separately, to explain the fundamentals of the DED process.²⁸ However, in most of the DED replicators, the deposition head consists of the laser and powder feeder inside the chamber, which makes the chamber larger and hinders mobility.

The design of our setup is different from the other cited works. By having the laser outside the processing chamber, the laser and powder feeder are controlled separately as opposed to a single deposition head. This enables us to have a hybrid system that has the capability to perform both DED and SLM. Furthermore, the smaller chamber makes efficient use of argon gas with less downtime for gas purging.

Since most studies have focused on synchrotron x-ray imaging during AM, which can tackle the DED challenge, our current effort is focused on the following: (i) understanding the AM process using x-ray diffraction²⁹ that can be used to approach the challenge (ii) in addition to (i). With the help of this custom instrumentation, we can track the solidification pathways in metallic alloys with high spatial and temporal resolution,³⁰ which can aid the prediction of microstructural evolution during AM. In addition, we can quantify the lattice strains as alloys evolve with temperature and time during the AM process. Finally, with this system, we can explore the effects of changing the laser scan mode, i.e., from laser rastering to spot weld (equivalent to a pulsed–laser-based AM), on material behavior.

The DED simulator was built at the Laboratory for Advanced Materials and Manufacturing (LAMM) at Cornell University (Fig. 1). The DED simulator can be separated into four parts: energy source, process chamber, stage system, and powder feeder delivery. The energy source is an air-cooled, 500 W continuous wave (CW) multi-mode laser from IPG Photonics (Model: YLR-MM-AC-500) [Figs. 1(a) and 1(g)]. It has a minimum focal point size of 0.5 mm (used in our experiments) and a wavelength of 1070 nm. The fiber diameter, the collimator lens focal length, and the focal length of the lens are 0.2, 100, and 250 mm, respectively. The laser is mounted on a structural support [Fig. 1(f)], with the working distance of the laser being 207.7 mm, focused on the powder bed surface or powder feed trajectory.

The process chamber [Fig. 1(e)] is 152.4 × 152.4 × 120.65 mm³ in dimension and is made of steel. The chamber contains a UV fused silica window (50 × 50 × 4 mm³) with a near-infrared protective coating. The windows where the x ray enters and exits are 47 × 19.3 mm² and 126.6 × 79.4 mm², respectively, and are made of 0.127 mm thick Kapton polyimide film, with >99% transparency to x rays (Du Pont Inc.) [Fig. 1(e)]. The size of the Kapton window was determined based on the dimensions of the diffraction cones generated by the first 10 (hkl) planes of Fe (BCC). In addition, there is a polycarbonate transparent glass window that can be used for thermal imaging integration. Outside of the chamber is an argon gas inlet [Fig. 1(d)] and, on another side, a check valve with a filter attachment. The x–y stage system consists of two motorized Thorlabs LNR50 stages that are controlled by BSC203 (Thorlabs) [Fig. 1(c)]. The process chamber is mounted on this staging system to allow for printing on the x-axis and y-axis with a motion range of 50 mm.

The powder feeder delivery channel contains a 5MPE feeder from Oerlikon Metco [Fig. 1(h)]. The carrier gas of choice is 60 psi of argon, along with 15–20 psi of air, which vibrates the hopper filled with powders [Figs. 1(d) and 1(i)]. The vibration due to air supply pressure also provides high flowability of powders through the nozzle. We used an angular directed nozzle whose angle could be changed anywhere between 0° to 90°. For this set of experiments, the angle was fixed at ∼60° from the x-axis. The nozzle is attached to the chamber via a telescopic joint [Fig. 1(j)], which is mounted on a Thorlabs y-axis motion stage [Fig. 1(b)]. This combined system provides additional degrees of freedom for moving in the x and z directions, which ensures that the focal point of the laser always intersects the powder stream during multi-layer printing.

The DED simulator was integrated at the Forming and Shaping Technology ID3A (FAST) beamline of the Cornell High Energy Synchrotron Source (CHESS) [Fig. 2(a)]. For the experiments presented here, high-energy monochromatic hard x rays with 61.3 keV energy, 0.202 Å wavelength, and a square cross section of 0.25 × 0.25 mm² are used in transmission mode. Mixed Mode Pixel Array Detector (MM-PAD) was used to capture portions of the diffraction cones.³¹ The q-coverage of the detector is 5.51–8.07 Å⁻¹ (q = $4πλsinθ$⁠) at the sample-to-detector distance of 640 mm. The MM-PAD consists of a 2 × 3 tiled array having a configuration of 256 × 384 active pixels, coupled to a 0.75 mm thick CdTe sensor. The pixel size is 0.15 mm × 0.15 mm, and it achieves full well depths of 4.6 × 10⁸ keV. All these features of the detector enable x-ray imaging to extend beyond 100 keV, with a maximum possible frame rate of 1.1 kHz.³² The lattice plane spacing values (d) are determined from the following relationship [Eq. (1)]:

where λ stands for the wavelength of the x rays and θ is the angle that satisfies Bragg’s diffraction conditions. It should be noted that we can use other detectors with our system; however, this paper is mainly focused on using MM-PAD detector because of its fast acquisition rate and high dynamic range. As demonstrated in Fig. 2(b), the detector is positioned ∼640 mm away from the sample and ∼139 mm above the x-ray beam. This was calibrated with CeO₂ standard powder on the substrate. With this detector configuration, the MM-PAD captures peaks along the given azimuthal range of 85°–95° [Fig. 1(h)]. For the given detector pixel size, detector position, and data analysis procedure, the strain uncertainty is estimated to be between 0.000 10 and 0.000 25, which corresponds to 0.1–0.25 times the strain resolution of a single pixel. The whole setup is mounted on a broad-board base attached to the Huber XY-Stage 5102.40 that provides additional degrees of freedom for movement. One of the unique features of this setup is the “effective canceling motion” such that the x ray always probes a fixed point with respect to the scan length. The Thorlabs stage moves the enclosure (referred to as primary motion), whereas the Huber stage moves in an equal and opposite direction (referred to as secondary motion). This allows us to probe a stationary point with the x-ray beam throughout the experiment. It should be noted that the laser and powder feeder is always stationary with respect to the enclosure movement. The combination of the y-axis motion stage and a telescopic joint for the powder feeder provides the necessary degrees of freedom along the x and z axes, respectively, for multi-layer prints.

The alloys used to demonstrate the capabilities of the DED simulator are Inconel 625 (IN625) and Stainless Steel 304 (SS304). IN625, a solid-solution, nickel-based superalloy, consists of niobium and molybdenum, which are the main strengthening elements in its Ni–Cr rich matrix. With a powder size range of 45–150 µm, the powders were produced via gas atomization from Steward Advanced Materials. SS304, an austenitic stainless steel often used for AM applications,³³ was obtained from Carpenter Additive with a powder size range of 15–45 µm.

Before the experiment begins, the chamber is purged with argon to maintain slightly positive pressure. This guarantees that the sample has minimum exposure to oxygen, while also removing any gas fumes or spatters, formed during printing, away from the melt pool. We print both powder bed based single line scans and powder fed DED single line scans. For the powder bed based single line scans, the bed has dimensions of 20 × 2 × 2.4 mm³ assembled on a stainless-steel substrate plate of dimensions 75 × 25 × 5 mm³. In addition, the bed is sandwiched between two 0.05 mm thick stainless steel shim tapes. The shim tapes were folded along their length to create powder bed walls. In our case, it provides enough structural integrity to hold the uniform powder bed and thermal boundary conditions closer to the industrial AM processing with respect to the more commonly used glassy carbon. The height of the shim tapes acts as the support to fully lay powders and then smoothen them out using blades. The intensity of the peaks from the shim tape is minimal or almost negligible as its relative volume fraction is very low (<5% of the total volume through which x rays are transmitted). Furthermore, the peaks of the shim tapes, made of 18-8 stainless steel, do not overlap with the peaks of IN625 that we have used for analysis. In addition, the unmelted powders between the shim tapes and the solidified alloy create a powder barrier, reducing the heat conduction to the shim tapes. Hence, our assumption is that the shim tape peaks do not change with time during laser processing.

For results shown in Fig. 3, the x-ray beam is located approximately between 500 µm and 1 mm above the substrate surface, where the x ray always probes a point at the center of the line scans. During the first second, the laser is off, such that the detector captures the powder diffraction patterns before melting occurs. As the x-ray beam slits open, the laser is turned on and scans a line measuring 14 mm. The MM-PAD captures peaks with a frame rate of 100 Hz, with a corresponding exposure time of 10 ms. The time it takes to complete the line scan is dependent on the speed of the laser. Once a line scan is finished, the laser is turned off, but the detector continues to capture the x-ray patterns of the solidifying material during a cooling time of 10 s. Along with powder-bed-based line scans and powder fed single line DED scans, spot weld experiments were also performed. Spot weld experiments consist of illuminating a fixed spot on the powder bed with the laser for a particular dwell time to study the effect of a symmetric laser beam profile.

In this section, we explain the various measurement modes that are possible using our custom DED simulator. We first present the data collected from the IN625 powder bed based line scan. It is known in the literature that the macroscopic stress state in IN625 correlates well with the (311) crystal plane in its γ-phase.^25,34 Hence, we focus on the (311) peak for most lattice strain and peak evolution data analyses.³⁵ The kymograph [as shown in Fig. 4(a)] is produced from the raw 2D x-ray evolution dataset. This analysis gives us a broad overview of the time-dependent dynamic nature of the AM process. To elucidate, the process can be divided into 4 basic stages, i.e., before melting, during melting or formation of the melt pool, during solidification in the mushy zone, and the solidified stage. Figure 4(b) shows the raw intensity vs 2θ of these four representative stages, including the unmelted powder layer, the melt pool, the solidification zone, and the solid cooling zone, respectively. In Fig. 4(c), we show a kymograph for spot weld of IN625 at a laser power of 150 W and dwell time of 0.95 s. Notice the difference in the shape by changing 2θ compared to Fig. 4(a). Here, it is apparent that the window for solidification of the spot weld is smaller [the orange box in Fig. 4(c)] than the former case of laser rastering [the orange box in Fig. 4(a)]. This is consistent with the shapes of the melt pool and mushy zone expected for the symmetric spot weld, as compared with the line scan, which is expected to have asymmetric dimensions with the moving laser. Such information gives us a fundamental understanding of the moving melt pool profiles and their effects on the solidification microstructure.

Two experimental conditions for powder bed based line scans of IN625 (keeping the speed constant), i.e., 150 W, 4.5 mm/s and 250 W, 4.5 mm/s were compared to show the differences in thermal history and lattice strains with time. The peak indexing was done by predicting peak positions (based on ICSD database for IN625) using nominal IN625 γ-phase lattice parameters. The fitting was done using a Pseudo-Voigt analytical function using the HEXRD module in Python. The stress-free lattice plane spacing for the FCC γ-phase of IN625 was obtained by averaging the lattice plane spacing of the bulk unmelted Pseudo-Voigt fit powder diffraction patterns recorded before every experiment. For the 150 W case [Fig. 5(a)], the energy density is smaller compared to 250 W [Fig. 5(b)]. This, in turn, changes the thermal history during the heating and cooling cycles. Both cases almost reach similar values of lattice strains at maximum heating, whereas, during the cooling cycle, the 250 W experiment has a smoother curve that reaches a steady state much earlier than the 150 W. This signifies different cooling paths in both the cases, which could be dependent on the different processing parameters. We also highlight two specific features from the lattice strain curves, i.e., significant fluctuations as observed especially for the 150 W curve and the regions of similar strains in both cases.

First, to explain the fluctuations, the reader is referred to Fig. 5(c). It shows raw stacked plots of intensity vs 2θ from t = 4.5 s–8.5 s for both experimental parameters. As observed from the black dashed lines [corresponding to the (311) peak], the peaks change considerably from t = 4.5 s–8.5 s for the case of 150 W, with respect to its unmelted powder diffraction pattern, having peak features such as asymmetry and splitting. Such changes may occur since at low energy densities, the heat input is not sufficient to fully melt the entire volume of powders probed by the x-ray beam. As a result, some unmelted powders remain and float around in the solidifying melt pool. These interactions lead to constant changes in orientations of the growing crystallites. However, as the melt pool cools and solidifies further, the fluctuations go away as the system reaches steady state. Second, to explain the similar strain region observed between the times ∼8.6 s and 10 s in the 150 W case, the reader is referred to Fig. 5(d). It shows six overlapping raw 1D integrated intensity vs 2θ curves between 8.6 and 9.1 s, with the inset showing a zoomed-in view of specifically the (311) peak. This signifies a dominating peak, growing in intensity, corresponding to either increasing crystallite size or rotation of that crystallite toward Bragg’s diffraction condition over time.

Figures 6(a) and 6(b) show the relative peak broadening for both the experimental conditions in their steady state. It should be noted that the unusual dips around 11.5° and 13.3° refer to the gap pixels in the MM-PAD detector. For Fig. 6(a), the raw 1D data are provided, and focusing on the (311) peak, we see that the 150 W peak (blue) is much shorter and broader as compared to the 250 W peak (orange). Therefore, the overall grain size in the 150 W case is expected to be smaller than the 250 W case. Also, it is possible to detect minority phases that are formed during solidification. An example is highlighted in Fig. 6(a), where a small peak beside the main (311) peak for the 250 W experimental case could point to the presence of a minority phase. However, whether the system detects such phases also depends on the volume fraction and the exposure times of the x-ray beam, coupled with the dynamic range of the detector. Figure 6(b) represents the integrated 1D intensity plots with peak fitting for the same experimental conditions as Fig. 6(a). Using the pseudo-Voigt fitting function as discussed above, we can quantify the full width at half maxima (FWHM), to estimate the crystallite size for both cases.

Figures 7(a) and 7(b) represent the powder fed single line DED measurements obtained for SS304. Figure 7(a) refers to one such diffraction spectrum in the steady state. We see that the 1D x-ray plot here shows different peak characteristics after solidification for DED as compared to powder bed based line scans discussed above [Figs. 5(c), 5(d), 6(a) and 6(b)], even with similar processing parameters. The peak splitting in these data originates from the diffraction pattern of unmelted powders (the green arrows) and textured solidified grains (red arrows). This is further verified by closer observation of the 2D x-ray plot in Fig. 7(b), where we see the spotty textured rings slightly (highlighted with red arrows) above the continuous rings (with a slight shift in 2θ). This could be attributed to the fact that with powder feeder in DED, some residual unmelted powders alongside the print are always present and depend on the catchment efficiency of the powders being melted by the laser.

We used our custom DED simulator to study (i) the evolution of diffraction patterns over time, (ii) differences in the melt pool solidification profiles with different experimental modes, i.e., laser rastering vs spot weld, (iii) the lattice strains with time describing the thermal history, and quantitative estimation of peak broadening, with examples across different processing conditions for the IN625 powder-bed-based line scans, and (iv) fundamental differences in the diffraction patterns for SS304 processed using powder fed DED, such as the appearance of both textured and continuous rings, as verified from both 1D and 2D diffraction plots. Changing the experimental modes helped us explore the capabilities of our simulator using two popular metallic alloys. X-ray diffraction provides complementary information on the material behavior and transient changes that occur along with the x-ray synchrotron imaging that is more widely employed. Deconvolution of the “encoded” x-ray data helps us understand different aspects of the complex AM process and further use of those data to verify numerical and computational models.

A.M. and A.D. gratefully acknowledge the funding received from the National Science Foundation CAREER Award No. CMMI-2046523 and the NASA University Student Research Challenge Award No. 80NSSC21K0465. This work was conducted at the Cornell High Energy Synchrotron Source (CHESS) facility, which is supported by the National Science Foundation under Award No. DMR-1829070. The authors also acknowledge Dr. Hugh Philipp and Dr. Mark W. Tate for assistance with setting up the MM-PAD detector for the operando experiments and other staff at CHESS for help during the beam time. We also thank Temesgen Worku for his assistance with designing the setup.

The authors have no conflicts to disclose.

The data that support the findings of this study are available from the corresponding author upon reasonable request.