Investigation of process by-products during the Selective Laser Melting of Ti6AL4V powder

This paper investigates the formation of process by-products during the laser processing of titanium alloy powders by Selective Laser Melting (SLM). The study was carried out during the printing of Ti6AL4V parts using a production scale SLM system (Renishaw RenAM500 M). By-product particles were obtained on the surface of powder removed from the area around where the pulsed laser powder treatments had been carried out. The process by-products examined in this study were damaged Ti6AL4V particles along with condensate. The par- ticles were found to exhibit deshelling, fracture, and collision damage. Based on TEM and SEM examination, the condensate particles were found to have sizes in the nanoscale range and exhibited morphologies, similar to those reported in the literature for welding condensates. Energy-dispersive X-ray spectroscopy (EDX) analysis indicated that the condensate formed from processing Ti6AL4V, exhibited a higher level of aluminum than that obtained for the alloy itself, lower levels of titanium with minimal vanadium levels, were also obtained. This may indicate that the alloy partially decomposes, with the emission of the lower melting point alloying element. The use of an in-situ melt pool monitoring system (called Renishaw InfiniAM Spectral), was evaluated for detecting the presence of these by-product particulates, based on photodiode measurements of the melt pool emissions, along with a camera-based imaging of visual per layer conditions. A reduction in the intensity of infrared emissions was detected, in areas where suspected spatter particles had been redeposited. Thus, de- monstrating that process monitoring can be used for the in-situ detection of particulate defects formed during printing.


Introduction
Powder consistency is a key consideration for the repeatability and predictability of components manufactured through the Selective Laser Melting (SLM), Additive Manufacturing (AM) process [1,2]. Moreover, the unused powder from a print is often recycled, however its' morphology and chemistry could potentially change due to the influence of by-products produced during printing, alongside chemical changes to the powder from oxidation/nitration [3]. The quality of the powder is particularly important for highly regulated AM applications in the medical device and aerospace sectors, where high powder costs and tight specification controls effect overall process costs [4]. In order to facilitate a higher level of SLM process quality control, the use of techniques to monitor the melt pool emissions are clearly important [5].
Several by-products are generated during SLM processing in Powder Bed Fusion (PBF), including ejected powder, spatter, and condensate as illustrated schematically in Fig. 1. Spatter is expelled from the melt pool during processing forming as new spherical particles of the alloy, which are noticeably bigger and typically chemically identical to the raw feedstock [6]. The chemical compositions of the spatter suggest that it is not generated by vaporisation. The spatter has several causes including higher absorptivity of powder over bulk, inconsistent powder feeding or layer thickness, and powder oxidation. Ladewig et al. outlined that ejected powder is an SLM phenomenon where particles of feedstock are ejected from the melt pool zone similar to welding plume dynamics [7]. Sutton et. al. investigated the by-products from the SLM processing of 304 L stainless steel, showing that the condensate was a mixture of all the alloying elements [8]. Wang et al. focused on the SLM printing of CoCr, with a number of spatter morphologies observed and reported [6].
It was found that the process gas flow, such as argon or nitrogen, acts to shield the processing zone from chemical reactions such as oxidation/nitration, whilst removing airborne by-products. Gas flow uniformity and velocity thus have a major influence on part quality, density, and surface morphology [7,9]. They also influence the condensation rates of the produced metal vapours, which can affect laser attenuation. The latter can occur, due to the diffraction of the beam arising from the presence of nano particles in its path. This may reduce the laser power, as the airborne by-products can absorb the laser energy and scatter the beam. Ferrar et al. and Kong et al. [9,10], observed this effect for SLM, the former noting that areas of low gas flow showed poor condensate removal and thus reduced mechanical performance. Kong similarly showed the influence of low gas flow on porosity generation within the printed specimen. Ladewig et al. [7] also reported that where low gas flowrates were used, the removal of the melt pool by-products was not efficient and redeposition was possible within the current or next layer, giving rise to poor printing conditions. Shcheglov et al. [11], reported on condensate particle generation and conducted plume attenuation measurements for high powered fibre laser welding of mild steel plates, with condensate particle diameters of 80-100 nm reported. Ladewig et al. [7], evaluated the influence of argon gas flow regimes on the generation of contaminates in the SLM processing of Inconel 718. Under 'unstable' gas flow conditions a range of processing by-products including condensate, spatter, and ejected powder were generated, along with poor printing results. It was concluded that optimisation of the argon flowrate is required to avoid the redeposition of these by-products. Leung at al. [12], showed that melt pool dynamics can be simulated, including the production of spatter as well as its' trajectories.
During SLM, the melt pool temperatures can attain values that are above the melting point of the alloy and possibly the boiling point of some alloying elements (Table 1). Ali et al. [14], modelled melt pool temperatures using Finite Element Analysis (FEA) to characterise temperature gradients globally within the specimen, and locally around the melt pool, examining both the melt pool depths and widths for varying processing parameters. Based on melt pool modelling and predicted peak melt pool temperatures for SLM processing of Ti6Al4V, Ansari et al. [15] concluded that they were in excess of the alloys melting point, which is typically in the range 1926-2226 °C. Such melt pool temperature regimes could lead to boiling, followed by vaporisation, of some alloying elements, with rapid condensation occurring when it contacts the process gas flow above the melt pool zone.
This paper investigates the condensate particles generated during the SLM processing of Ti6AL4V, which due to its material properties is widely used for the printing of medical device and aerospace components [16]. The focus of this paper is firstly the evaluation of the type of process by-products such as ejected powder, spatter, and condensate, obtained during the SLM printing of Ti6Al4V. A second objective is to evaluate the use of an in-situ optical emission monitoring system (Renishaw InfiniAM Spectral), for the detection of these by-products. While a number of researchers have highlighted the benefits of process monitoring for quality control purposes in metal additive [17][18][19][20]5] this is the first know evaluation of an in-situ analysis technique for the identification of by-products generated during part printing by SLM.

Experimental design
Printing studies were carried out using a Renishaw RenAM500 M SLM Powder Bed Fusion system, which is equipped with a 500 W laser (λ =1.07 μm), with an in focused spot size diameter of approx. 75 μm. The RenAM500M system can operate in both continuous laser mode and in a modulated mode. In the latter case, the laser fires for a given amount of time after which it switches off and moves a defined distance to the next point location, before firing again for a given time, known as the exposure time. During this study all test specimens were created using the modulated laser mode. The AM equipment's build volume is 250 mm × 250 mm × 350 mm. For the examination of the by-products generated during SLM processing of Ti6Al4V a simulated production build was developed using an array of 117 flat specimens with dimensions of 6 mm × 3 mm × 100 mm. These represented the printing of a small cross section area part, evenly placed across the build plate. The specimens were printed using the following conditions: layer height: 60 μm, power: 200 W, point distance: 60 μm, and exposure time: 70 μs, which are the printing parameters recommended by the original equipment manufacturer. Powder samples were collected for examination taken from the surface of the powder bed on the left and right of the printed specimens, with condensate collected from the chamber. The process gas flow, argon in this case, flows over the build plate from right to left (when looking in from the front of the machine), designed to remove by-products, with a small percentage redeposited to the left of the specimen. The resulting Ti6Al4V specimens exhibit the following properties: density of ∼99.8 %, pores of ∼20−300 μm and a microstructure dominated by martensite needles, typical of the printing process. These properties were determined using computed tomography  measurements (Phoenix Nanotom system), as well as based on microscopy examination of parts which were sectioned and polished, followed by etching using Krolls reagent.
Process monitoring was carried out using the Renishaw InfiniAM Spectral system [21]. This facilitates the measurement of the optical emissions from both the laser and melt pool using photodiodes. The data is recorded and then reconstructed into both 2D and 3D views, in near real time. As shown schematically in Fig. 2a 500 W ytterbium laser (1) emits light with a wavelength of approximately 1070 nm. A small amount of the laser passes through a fixed mirror and is detected in the laser monitoring (LaserView) module (10), this module provides information about the laser input energy, at a rate of 100 kHz.
As the laser interacts with the powder, a melt pool (18) is created. From this, a proportion of the emitted radiation (13, 15 & 17) is transmitted back up the laser path, where they are directed towards two photodiodes (4 & 5). These photodiodes make up the melt pool monitoring (MeltView) system (2), which also records data at a rate of 100 kHz. The co-axial nature of the monitoring system results in a field of view that is determined by the laser scanning system. This type of coaxial photodiode PM system is similar to that studied in [23] and [5].
A custom software was developed in C++ to filter and analyse the data from both MeltView and LaserView, focused on the near-infrared emissions. This software cuts the data from a specified region, analysing it for average, sum, variance, and standard deviation. Excel was used to graph the analysis of data from MeltView, LaserView and ImageJ, as well for linear regression analysis. The RenAM500 M also has an inbuilt camera to record a 1000 × 1000 pixels image before and after each layer, typically used for comparison of powder distribution and detection of obviously defective layers. For this paper an ImageJ macro cropped the images to focus specifically on a specimen and then threshold it for detection of minor defects of perhaps a few pixels. These defects were areas where suspected redeposition of by-products occurred. ImageJ was also used to perform histogram analysis on specific Regions of Interest (ROIs) to understand the area of a region affect by a potential defect.

Ti6Al4V powder
The powder used for this study was obtained from AP&C powders, D90 < 43 μm Grade 23 Extra Low Interstitial (ELI) Ti6Al4V. The RenAM500 M uses a powder recycling system, therefore the powder used in this study had been processed within this recycling system over ∼9 months, along with the use of a top-up regime. Just to stress that over this ∼9-month period, the AM equipment was used relatively infrequently, compared to that normally expected for commercial AM processing operations. The recycling system includes the use of a 63 μm ultrasonic sieve and cyclonic separators design to remove powder from the argon gas flow used to transport the powder around the machine.
For the collection of used powder, the samples were taken from around the printed samples at a number of locations across the build plate. A clean stainless-steel scoop was used to collect the powder samples. These powder samples were then stored under an argon atmosphere, in a glass vial.

Characterisation
Powder samples were examined by mounting them on carbon tape to facilitate SEM / Focused Ion Beam (FIB) analysis, with the samples taken in the centreline of the specimen array, i.e. halfway between front and back of the build plate and left and right of the specimen. A Zeiss ULTRA plus SEM with a 20 mm² Oxford Inca Energy-dispersive X-ray spectroscopy (EDX) detector was used for powder morphology analysis. SEM Analysis was performed at 10KeV. A FEI Quanta 3D FEG with FIB milling capabilities was used to examine representative samples of particles for inspection of their internal morphology. The powder analysis was conducted on a subcontractor basis by Glantreo Limited, Co. Cork, Ireland. Inductively Coupled Plasma (ICP) was used for general elemental analysis, Inert Gas Fusion (IGF) for H, N, O measurements, and Carbon infra-red (CIR) for carbon content. The morphology and particle size of the condensate on a nanoscale was examined using a Transmission Electron Microscopy (TEM) FEI Titan 300 and a Hitachi Regulus 8230 SEM. For TEM the condensate particles were drop-cast onto ultrathin carbon TEM grid and imaged in conventional bright field transmission mode at 300 keV.

Condensate sample preparation
Visual examination of the powder bed after printing, demonstrated the presence of darker condensate particles, compared with that of the metal alloy powder. The condensate was collected using a lens cleaning swab from the left-hand side of the machine's chamber, which is the opposite side to that of the Argon purge gas inlet. Isopropanol Alcohol (IPA) was used to wash this swab and the condensate mixture was collected in an 8 mL glass vial. Due to its very small particle size, the condensate was found to be suspended in the IPA, in contrast the relatively larger Ti6Al4V particles picked up by the lens cleaning swab, were found to settle to the bottom of the vial. A pipette was used to precisely drop the suspension of condensate in IPA onto a copper substrate holder for SEM analysis. The alcohol was then evaporated by placing this substrate on a hotplate at 60 °C leaving a layer of condensate. For TEM analysis the suspension was drop cast onto an ultrathin carbon TEM grid.

Results
The first step in this investigation is the examination of the Ti6Al4V feedstock powder in order to determine if it was in compliance with ASTM Grade 23 [24]. The morphology of both the virgin powder and that obtained after printing (used powder) were evaluated. The trajectories of spatter material were examined, in order to describe the spatter's behaviour during processing. A summarisation of the morphologies of the produced by-products was determined, with FIB cross sections obtained for investigation of internal morphology. The morphology and chemistry of the collected condensate samples are presented, with results on both particle size analysis and composition reported. Finally, the analysis of the melt pool emissions in relation to print bed regions where by-products were and weren't identified, is reported.

Powder analysis
The analysis results of the powder feedstock are presented in Table 2. As highlighted earlier the used powders given in this table, are representative of powder condition after ∼9 months of use in the Renishaw's system, in which a top-up regime is employed, whereby the Fig. 2. Schematic of the InfiniAM Spectral monitoring system used in this study [22].
powder is never removed from the system. The powder was stored under argon within the AM system and when transported was also held in an argon atmosphere. Table 2 demonstrates that the used powder exhibits a very similar chemical composition to that of the as-received (virgin) powder. These analysis results demonstrate that the Ti6Al4V feedstock powder is within the specification for Grade 23 [22].

Spatter generation
A video was obtained using a 16 M P video camera of the generation of by-products from SLM processing of Ti6Al4V, in order to obtain an understanding of the spatter generation regime. As demonstrated in Fig. 3, the spatter particles are generally diverted to the left of the processing zone, by the argon gas flow. Visual examination of the still images in Fig. 3 indicates that the spatter exits the melt pool towards the previously formed weld track. In the situation where the melt pool formation or flow direction is opposite to the gas flow (top left image), the spatter exits the processing zone in a linear trajectory, aligned with the gas flow. The top right image in Fig. 3, illustrates a case in which the melt pool flow is in the same direction as the argon flow. In this case the spatter's trajectory initially goes against the gas flow direction, however it is then redirected, towards the left.

Powder morphologies
The typical particle morphology of the virgin Ti6Al4V powder is shown in Fig. 4A, it exhibits a highly spherical form and some surface cracks/grooves, with depth of approximately a few nm. The morphologies of by-product powder particles consisted of deshelled, fracture, and fused morphologies as shown in Fig. 4B-D. These morphologies are Table 2 Averaged ICP, IGF, and CIR analysis results of virgin powder and used powder after 9 months taken from the Renishaw RenAM500 M using a powder top up regime and closed loop powder sieving and recycling.  typical of those previously reported in the literature for SLM [6]. It is important to note that the virgin powder was not found to exhibit a "shell" structure, as demonstrated by the FIB cross-section analysis presented in Fig. 5 (left). It is hypothesised that thermal cycling of the powder particles during the processing results in thermal stresses, which promote crack propagation. This may explain as to why the cracking is confined to a local area on the particle surface and was not observed within the particle bulk.
In the case of the fused morphology, two types were identified. One is associated with a molten particle colliding with a larger particle and solidifying in a random fashion around the larger particle. This is seen in Fig. 4C (left) and it appears the small particles, not fully molten, collide with a larger particle that is close to molten. When these particles collided, the smaller particles appear to be partially bond to the larger particle. It is also possible that the larger particle becomes molten and due to surface tension forces is attracted to the smaller particles and hence engulfs them partially. FIB analysis was used to examine a representative example of fused morphology, as shown in Fig. 5 (right), indicating what appeared to be adhesive bonding between these particles. As illustrated in the figure the gap between the particles may be partially filled with fragments of smaller particles.
The fractured morphology shown in Fig. 4D, appear to have been    damaged during the laser treatment. From the SEM/FIB examination these fractured particles are characterised by the loss of a significant proportion of the assumed original spherical volume, along with a relatively rough new surface created. These newly formed surfaces are not the typically smooth spherical surface characteristics of a re-melt and solidification, which occurs due to the surface tension of the molten alloy. From this observation it is hypothesised that the particle has been laser cut during processing and ejected rapidly from the plasma plume or melt pool zone, before there is sufficient time for it to be fully melted, as such it is considered to be a form of ejected powder.

Condensate analysis
As detailed in the literature review, it had been proposed that the melt pool temperature is sufficiently high, to vaporise the alloying element of Ti6Al4V [15]. The results of the EDX analysis of the extracted condensate particles, are given in Table 3. The first thing to notice about these results is the very high level of carbon, this is likely to be largely associated with hydrocarbon residue obtained after the evaporated IPA solvent, used in the extraction of the condensate. Of particular interest from these results, is the relative ratios of the Al, Ti, and V elements. Compared with the elemental concentrations in the alloy, which are also included in this table, the condensate exhibits a significantly higher concentration of Al. This result indicates that there is a dissociation of the alloy when it is vaporised, and that the generated condensate incorporates much higher levels of the aluminium, which is of the lowest melting point alloying element.
SEM and TEM images of the solvent extracted condensate sample are given in Fig. 6. The SEM image demonstrated the agglomerated nature of the particles extracted using the solvent, while the TEM images demonstrate the nanoscale particle size of an individual condensate particles within the small, isolated aggregated clusters. From the TEM analysis of the condensate the particles are seen to be spherical in an agglomeration, with particle size typically in the range of 5-20 nm.

In-situ process monitoring
From the image-based process monitoring of the powder bed during printing, it was observed that several small dark regions appeared on the surface of the printed layer, which appeared based on microscopy examination to be associated with discontinuities in the melt pool track (Fig. 7 (Top)). These discontinuities appeared as both peaks and troughs, visually indicating potentially trapped spatter, or a by-product redeposit. During the SLM process by-products such as spatter, condensate, and ejected powder can redeposit onto the printed layer or powder bed. These discontinuities or darkened regions are considered linked to a by-product redeposit in the previous layer, as demonstrated in Fig. 7. An experiment was conducted to investigate the use of the InfiniAM Spectral melt pool emissions for the monitoring of such regions. The objective was to evaluate if these suspected redeposits, had an influence on the melt pool emissions as monitored using the In-finiAM system. Typically, the melt pool data is examined via Renishaw InfiniAM Spectral software, however in this case it was determined to be inconclusive for detecting such small anomalies efficiently. Instead it was found necessary to investigate the use of the primary data signals obtained from the photodiodes used by the InfiniAM Spectral system.
A rectangular block sample with a constant layer of 35 mm × 20 mm was printed, with the hatch lines aligned to the gas flow. Within this block, consisting of 506 layers, a set of 60 layers were selected for detailed investigation. Within each of these layers, four regions of approximately 2 mm × 2 mm were examined, two of which exhibited a darkening of the surface (labelled R1 and R2) and two which exhibited normal brightness and thus acted as the control (labelled R3 and R4). Any region which exhibited a darkness of less 59 on a brightness scale of 255 was considered to be a defect by ImageJ (thresholding method), as seen in Fig. 7 (left), with the four selected regions locations indicated here also. In this case anything that is below the threshold is determined as a potential defect and indicated in red, with anything that's not a defect indicated in yellow. A histogram analysis was used to determine how many pixels of red and yellow were in each region for each layer. From this, the percentage area covered by red pixels, i.e. a potential defect, was calculated for each region for each layer, referred herein as the percentage area coverage. The trend of the percentage area coverage was then plotted for each region versus the layer number to determine the trends. It was seen that a trend of very high levels of potential defects in R1 and R2 existed, with regions R3 and R4 exhibiting relatively low defect trends. The recorded near infrared emission data was filtered to look specifically at the four selected regions, R1, R2, R3, R4, with the average infrared emissions in the region for each layer calculated. These averages per layer were then normalised relative to the average emissions over the 60-layer range and the minor laser variations compensated for. Fig. 8 shows the near infrared emissions plotted against the layer number for regions 1 and 4, with the inverse trend of the percentage area coverage for the corresponding layer also shown, which has been scaled at 125 % for visualisation of correlation between the image-based analysis and the near infrared emissions. The reason the inverse is used in this case is that as the percentage area coverage goes up, the near infrared emissions reduce, hence taking the inverse allows us to see this more clearly. For the region R1, the graph shown in Fig. 8 shows there is a strong correlation between the trends seen from the near infrared emissions and the percentage area coverage. This near infrared emission intensity trend was also demonstrated for region R2. A linear regression analysis was performed on the melt pool emissions shown in Fig. 8. The R 2 value for R1 is significantly higher than for R4. It is seen that the trend for region R4 is relatively consistent with little variation in the emissions recorded. This trend was repeated for region R3. By way of providing a perspective, the spikes seen in R4 would be maximum ∼2 % of the photodiodes' range, hence a small variation. Regions R3 and R4, which were areas with little redeposits or melt pool track defects, demonstrated the signal noise that exists naturally within the process, excluding laser variation, is influenced by layer height inconsistency, powder chemistry, as well as morphology.
It is hypothesised that the near infrared emissions are lower when the laser is exposed to a solid Ti6Al4V layer, as opposed to the powder, the heat generated by the laser is more easily conducted through a solid compared with the powder. The powder has a packing density roughly half of the printed material reducing its efficiency for conduction. In this case a reduction of the near infrared emissions is seen where a potential defect is more prominent in regions R1 and R2, further supporting the hypothesis that the laser is firing onto a redeposited spatter. Given that the spatter exits the melt pool in a molten state it can bond to the previous printed layer's surface when redeposited. In this case if the laser fires onto this piece of spatter during the next layer it is easier for the laser's energy to be conducted into the specimen as opposed to radiated out of the melt pool, hence why a reduction in infrared emissions would be seen.
Using data combined from both the camera-based in-process images and melt pool emission data, provides a more accurate approach to monitoring the melt pool. The process monitoring results indicate that there is a correlation between the melt pool track defects, which are suspected to be linked to by-product, and the intensity of the near infrared emissions. The ImageJ based image analysis is an efficient, scalable and repeatable method, removing any manual errors from the analysis. There is a correlation between the trends seen in the inverse of the area percentage coverage in a region and the near infrared emissions.

Conclusion
This paper investigated powder by-products produced during SLM pulsed laser processing of Ti6Al4V and their detection using the Renishaw InfiniAM Spectral process monitoring system. It was concluded based on an examination of powder, which had been used in a production scale AM equipment (RenAM500 M) for approx. 40 print runs employing a top-up regime for recycling, that: • By-product powder morphologies have been observed, characterised, and investigated with three types of particle morphology identified; deshelled, fracture, and fused morphology.
• TEM analysis showed that the extracted condensate particles consisted of an agglomeration of individual particles with typical diameters of 10−20 nm. EDX examination indicated higher concentrations of Al and lower concentrations of V and Ti compared with that obtained in the original Ti6Al4V alloy. This indicates that the condensate is formed from the dissociation of the metal alloy with the loss of aluminium, which is the lower melting point alloying element.
• Investigation of melt pool emissions in relation to suspected SLM byproduct redeposits has been undertaken with links seen between regions of high redepositions or defected melt pool tracks and infrared emissions. The performance of the hybrid process monitoring combining the camera-based images and melt pool emission data was successful in detecting anomalies and linking them to a suspected by-product redeposit. This indicates the potential of using the primary data signals obtained from the InfiniAM Spectral photodiode detectors for melt pool quality control and analysis during part printing to potenially track component defects.

Author disclosure statement
No competing financial interests exist.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.