Electron-optical in-situ crack monitoring during electron beam powder bed fusion of the Ni-Base superalloy CMSX-4

Electron beam powder bed fusion (PBF-EB) belongs to the powder bed-based group of additive manufacturing (AM) technology. Using an electron beam as the heat source offers significant advantages compared to similar AM technologies, such as selective laser melting (PBF-LM). Unlike in PBF-LM, beam deflection is not done using mechanical mirrors but electromagnetic coils [1]. The beam deflection is, therefore, almost inertia-free and allows velocities of up to 10⁵ m/s on the building area. Furthermore, the electron beam can be quickly focused or defocused to control the energy input into the powder bed. This allows the development of novel process strategies for demanding materials, such as Ni-base superalloys [2].

CMSX-4 is a second generation Ni-base superalloy which was originally designed for the casting of components for high-temperature applications, such as turbine blades [3]. Nevertheless, gas-atomized powder composed of CMSX-4 has also been successfully processed into dense cuboidel or cylindrical samples with diameters in range of 12–23 mm and heights of up to 80 mm. The resulting microstructure varies up to single-crystalline samples depending on the applied process strategy. [4‐7].

In general, enhanced in-process monitoring has the potential to further establish powder bed-based AM of Ni-base materials thanks to the improved process controlability as well as understanding of defect generation [8]. Conventional PBF-EB machines only offer optical process observation (i.e., IR-cameras) through viewing windows. However, these viewing windows become darkened by metallization over the course of a build experiment. The source of metallization is the evaporation of volatile elements from the generated melt pool [9]. Either mechanical shutters [10] or a rolling film [11, 12] in front of the viewing windows have been suggested by literature as remedial measures.

Arnold et al. demonstrated electron-optical (ELO) imaging as a robust alternative for process observation during PBF-EB. Similar to scanning electron microscopy, backscattered electrons (BSE) are detected for in-situ imaging of the build area [9]. Its versatility for in-situ quality control during PBF-EB has been shown by different researchers, i.e., for fast process window development by Pobel et al. [13], validation of melt pool simulations by Breuning et al. [14], in-operando monitoring of the melting and dimensional accuracy by Arnold et al. [15, 16], for an arbitrary geometry with complex features by Ledford et al. [17] or off-axially arranged dual-detector ELO-imaging by Zhao et al. [18]. Wong et al. even modified the heat shields of a commercial PBF-EB-machine for ELO-imaging [19]. An overview of the expanding research field of in-situ monitoring is given by Grasso et al. in a recently published review paper [20].

This work exemplifies ELO monitoring during PBF-EB under the demanding process environment of the Ni-base superalloy CMSX-4. A freely programmable PBF-EB machine was combined with an integrated four-segment ELO detector and in-house developed imaging software to monitor the topography and sample cracking. Different imaging modes were compared to the corresponding sample surfaces and the robustness of the system was utilized for monitoring surface cracking over the entire build time.

For the experimental part, a freely programmable PBF-EB system “Freemelt ONE” (FM1, Freemelt AB, Mölndal, Schweden) operating at 60 kV acceleration voltage and up to 6 kW of electron beam power was utilized [21, 22]. The conducted calibration of the FM1 system resulted in a minimum beam diameter of 250 µm FWHP (= Full Width Half Power). The used CMSX-4 powder was argon gas-atomized (TLS Technik GmbH & Co. Spezialpulver KG, Bitterfeld, Germany) from CMSX-4 feedstock rods (Ross & Catherall Ltd. Sheffield, UK). Only the fraction of 45–105 µm was used for this work.

The system was equipped with a four-segment electron detector developed by Freemelt AB and in-house developed ELO imaging software. The build cycle for each layer was based on a regular PBF-EB cycle as depicted elsewhere [2]. After raking, a preheating step with a defocused electron beam was conducted to prevent powder smoke. After all subsequent melting steps had been finished an additional scan step for ELO image acquisition was added before the next powder spread by the rake. Hence, this modified PBF-EB cycle was identical to the cycle illustrated by Arnold et al. in [9].

The four-segment ELO detector was centred around the electron gun opening of the vacuum chamber at 165 mm above the build area. Therefore, the build direction BD is parallel to the z-axis and perpendicular to x-axis and y-axis in all ELO images presented in this study. An imaging area of 60 mm × 60 mm was scanned line per line with a scan line-distance l_off,ELO of 0.10 mm, a beam power P_ELO of 60 W and a beam deflection velocity v_scan,ELO of 10 m/s. The amplifier of the detector system operated at a sampling rate of 250 kHz. A python-based ELO imaging software was in-house developed specifically for image generation with the utilized four-segment detector. The ELO images were generated by logging the intensity signal of each detector segment to the corresponding beam position during the ELO scan. This was done for all detector segments individually as well as for combinations of detector segments. In addition, subtracting the individual imaging data of a pair of opposite detector segments was utilized for imaging the topography of the melt areas, whereas adding the imaging data of these two segments was utilized for imaging the material contrast. Furthermore, a mean image was generated by averaging the imaging data of all four detector segments.

For the build experiment a polycrystalline IN718 start plate with a diameter of 90 mm and a thickness of 22 mm was used. The build was started at a vacuum level of < 5 × 10^–4 mbar and after the plate was heated to 1050 °C. Temperature measurement was done using a thermocouple attached to the bottom side of the start plate. Before each new powder rake, the build tank was lowered by 50 µm, whereas the powder tank was lifted by 150 µm. 9 cubes with 15 × 15 mm² layer surface and 25 mm of height were built. An inverted 45° pyramid with a height of 7 mm and an upper surface of 15 × 15 mm² was used as a connection between each sample and the start plate. Thin support pillars were molten around and in between the actual samples to support powder attachment to the plate in the first few millimetres of build height. Altogether, the given build experiment consisted of 640 layers of 50 µm height each. For all samples, snake hatching as well as a 90° rotation of the scan direction with each new layer was applied over the entire build height. Besides, no contouring was used in this work.

Process parameters were applied to each cube individually with beam power P_melt in the range of 300–488 W, hatch line distances l_off,melt of 0.10 mm and 0.15 mm and beam deflection velocities v_scan,melt of 400 mm/s and 750 mm/s. The parameter range was chosen from previous studies [5, 6] to ensure dense samples without surface uplifting, which may otherwise cause significant damage to the rake system and, therefore, require the premature stop of the experiment.

Layerwise ELO imaging of the entire PBF-EB experiment was successful despite the increasing metallization on all four detector segments. Figure 1 shows the upmost layer 640 at the final height position of 32 mm utilizing different imaging modes. Depending on the position of the detector segments, a pronounced brightness gradient in the generated images for individual (Fig. 1a) or two combined detector segments (Fig. 1c, ) is observable. The dual detector ELO-experiments by Zhao et al. show a similar gradient in the image brightness of single detector imaging as well as the topographic imaging [18]. In the present experiment, only the mean image of all four detector segments (Fig. 1b) does not show this effect.

Figure 2 shows a close-up comparison of the surfaces of a cracked and a non-cracked sample using the four available ELO imaging types as well as their post-process counterparts. A surface-spanning crack on the sample surface (Fig. 2e) can also be identified in the single detector image (Fig. 2a) as well as the mean image (Fig. 2b) and the material contrast image (Fig. 2c), yet not in the topography image (Fig. 2d). By comparison, the topography image is best suited for the depiction of surface features, such as uplifted centres and pits. Thus, the large black dots visible in the other three imaging types (Fig. 2a, b, c) can be identified as spherical foreign bodies lying on the previously molten surfaces. As indicated by the dashed circles, no pores or surface pits of equivalent size are present at the same positions on the sample surface (Fig. 2e) after freeing the samples from the sintered powder bed. A potential origin of these particles is sputtering during the melting step.

Figure 3 illustrates how layerwise ELO imaging can consequently be used for monitoring cracks in melt surfaces of CMSX-4 over process time. Figure 3a shows the ELO mean imaging of the last molten layer 640 at a height position of 32 mm after roughly 30 h of build time. Five out of nine melt areas (#1, 5, 6, 8, 9) show surface spanning cracks. The comparably long process time is caused by the extended heating time for each layer to balance the built temperature.

In Fig. 3b, the evolution of the temperature at the bottom of the start plate is plotted against the build time and the build height. After the build start at > 1050 °C, the build temperature was stabilized at about 1030 °C. The subsequent decrease by 2.6 °C per millimetre build height is likely caused by the increasing distance between the thermocouple and the molten surfaces. As indicated, all cracked samples in Fig. 3a were crack-free until at-least 16.50 mm of build height or, respectively, 15:14 h:min of build time.

Figure 3c exemplarily shows the appearance of surface cracking for sample #6 using close-ups of the generated mean images. The cracking was first observed at the height position of 20.00 mm (18:33 h:min, layer 400) and was already surface spanning yet poorly visible in comparison. Until 24.00 mm (22:18 h:min, layer 480) it becomes more pronounced with a cross-like appearance. This is most likely caused by increasing width of the crack. However, its cross-like shape is not present anymore in the ELO images at 30.00 mm build height (28:02 h:min, layer 600). It is also not present in the last ELO mean-image at 32 mm height (29:55 h:min, build end) as well as afterward on the sample surface. However, some ELO images of layers in between 30.00 mm and 32.00 mm of build height still show a more cross-like shape. Overall, the comparison of layerwise ELO-imaging and the sample surface indicates the presence of a large crack network inside the sample. In general, these surface-spanning cracks are in the middle of the melt surface and oriented parallel to the lateral direction. The discussion of the formation mechanism and the nature of theses cracks is beyond the scope of this study. However, in-situ ELO monitoring offers a practical approach to investigate the development of cracks during build process.

A freely programmable PBF-EB system with an integrated four-segment ELO detector and in-house developed ELO imaging software was successfully applied for the in-situ crack monitoring during PBF-EB of CMSX-4. Comparing ELO imaging with the respective sample surfaces showed identical surface-spanning cracks. Using the layerwise monitoring allows to correlate crack initiation with the build height or the corresponding layer number. The conducted experiment clearly demonstrates the robustness and versatility of a four-segment ELO imaging system as a build quality control tool for PBF-EB.