Using X-ray Fluorescence to detect contaminants in metal powders used for additive manufacturing

Introduction

Despite the potentially damaging impact of contaminants on properties of AM parts, detecting undesired intruders in metallic powder feedstocks remains a challenging exercise. There is consensus on the definition of contamination, but no widely accepted strategy on how to detect in a non-destructive, cost-effective way directly on powders (see ISO/ASTM 52907 [1]).

Contamination can be introduced into metal powder even during the atomization process. Further, it may occur during powder handling operations linked to powder re-use, such as wearing and damaging of processing equipment, bad housekeeping practices and poor powder management.

Many organizations have developed their own practices and assessment criteria for powder contamination using a range of available test methods. The most popular ones are based on screening the differences in structure and chemistry [2, 3]. The portfolio of analytical techniques ranges from (i) optical microscopy, (ii) scanning electron microscopy coupled with energy dispersive X-ray analysis, (iii) X-ray photoelectron spectroscopy, (iv) X-ray diffraction, and going to the more sophisticated (iv) X-ray computed tomography (CT).

Almost a decade ago, Slotwinski and Garboczi [4] reviewed the metrology challenges linked to characterization of AM feedstocks and the considerable gap in standardization for conducting interlaboratory studies on such materials.

Without referring specifically to contamination, the guide published by ASTM committee F42 under the designation F3049 [5], advises which properties of metallic powders should be scrutinized before introducing the feedstock into an AM process. The major categories for powder characterization include (i) size distribution, (ii) particle morphology (shape), (iii) chemical composition, (iv) powder flow properties and (v) powder bed density.

In the more recent standard for assessing the quality of AM feedstocks: ISO/ASTM 52907 [1], evaluation of contamination is briefly mentioned. Unfortunately, the information is rather limited, and the end user is advised to perform either a microscopy study or use any “alternative testing practice, subjected to a prior customer/ supplier agreement”. This standard also mentions X-ray fluorescence as one of the analytical techniques suitable for chemical characterization.

Why XRF is a great investigative tool 

XRF has excellent potential as an investigative tool to identify different contamination types in metallic powder feedstocks. When a sample is exposed to X-rays, the response is a unique set of signals called characteristic fluorescence lines which are the chemical fingerprint of the analyzed material. Intensities of fluorescent signals can be correlated to concentrations. 

Quantitative XRF analysis is regularly achieved with calibrations based on matrix-matching standards. If samples belong to a certain alloy type it is expected that the same physical phenomena (i.e., radiation absorption/enhancement) will govern the final XRF response. Hence calibration lines provide the relevant correlation between the intensities of selected, alloy-specific fluorescent lines and the concentration of each analyte.

In qualitative XRF, the focus is identification of all elements present, with a secondary aim of giving concentration estimates. If samples are completely homogenous at micro-scale, even qualitative XRF investigations generate good compositional data.

Regarding the sensitivity of fluorescence to elements with different atomic numbers, the technique can sense if a specimen contains significant amounts of any element with Z > 11 (sodium). For detecting ultralight elements (like nitrogen or oxygen), XRF is not the preferred technique.

It is understandable why X-ray fluorescence is gaining attention from the AM field. The technique is extensively used in upstream metal production. Traditional manufacturers of metal alloys follow guidelines and restrictions of ASTM test methods for wrought/solid metallic alloys (i.e. E539, E572, E2465) [6-8]. These standards have been developed exclusively for wavelength dispersive WD-XRF spectrometers. The main aim of such investigations is to make accurate measurements of elemental concentrations. Straightforward application of these test methods in the AM context would mean first printing then polishing to a flat surface, followed by measurement. The high local temperatures generated at printing locations may enhance volatility of some metals and the printed part would not reflect entirely the composition of the starting feedstock.

A major advantage of XRF spectrometry is the relatively short measurement time (usually less than 10 min). For metallic powders in particular, the time required for making the specimen is also extremely short (less than 2 min). Another benefit of XRF is that, after measurement, the powder can be recovered and either subjected to additional tests or reintroduced into production. If price and equipment dimensions are important decision factors in the purchase process, then benchtop ED spectrometers may be much more attractive.

In this whitepaper, we explore the potential of EDXRF for surveying the composition of various metallic powder feedstocks in virgin and contaminated states. The benefits and constraints of the technique are discussed, with the intention to stimulate a wider audience to use the technique and to execute and interpret the measurements correctly.

BENCHTOP SOLUTION 50kV benchtop EDXRF spectrometer Robust and compact construction Proprietary X-ray tube Large area SDD detector Sample spinning (representative average)  Ideal for elemental screening Enlarged inspection area of the sample when using 9430 042 31221 sample holders 9430 500 00671 -P2 (large) sample cups

Samples and measurements

In this study two main categories of AM powders have been investigated: (a) virgin feedstocks - from different commercial sources, including the IARM certified referenced powders, and (b) deliberately contaminated test samples with customized contaminant concentrations. These special mixtures have been prepared at MTC - Coventry, UK.  In the latter case, controlled cross-contamination between two powders was introduced in terms of volume percentage. To simulate real-life situations, the levels of added contaminants were kept very low, starting from 100 ppmv (parts per million in volume). The powders used belong to different alloy types: titanium alloy Ti-6Al-4V (Ti64), stainless steel SS 316 L, nickel alloy Inconel 718, aluminum alloy AlSi10Mg and pure tungsten W particles. The particle size distributions and average particle size per alloy type are all typical for powders used in laser powder bed fusion applications.

All measurements in this study were performed on loose powders. Aliquots of powders can be poured directly into disposable analysis cups, assembled with a supporting (bottom) polymer foil. 

The X-ray signals excited in the specimens were measured using a Malvern Panalytical Epsilon 4 – benchtop EDXRF spectrometer, equipped with a 15W silver anode X-ray tube, 6 software-selectable filters, a high-resolution SDD30 silicon drift detector and a 10 - position sample carrousel. 

The Epsilon 4 also has flexible voltage and current settings that can be used to define application-specific excitation conditions for different metals. The measuring geometry is also optimized to maximize the area exposed to X-rays, and the continuous rotation of the sample during measurements gives better average information on the sample’s composition. 

What does the data tell us?

Most of us are familiar with X-rays used in medical environments or at border custom-control which uses X-ray transmission imaging. That creates the preconception that X-rays will penetrate very deep into the material and that what is measured gives a representative response for the entire body of the analyzed specimen. People accustomed to XRF analysis would tell you that what is actually measured strongly depends on the absorption of radiation inside the material. If samples are liquid (i.e., low absorption coefficients) then fluorescent signals would give a relatively good representation of the bulk. For metallic samples, where the absorption coefficients are high, detectable X-ray signals are from the first few micrometers (measured from the surface that is closest to the X-ray tube) up to several hundreds of micrometers inside the sample. That would not be a problem if the samples are perfectly flat and homogeneous at microscale. In this respect, powder analysis using XRF presents some challenges.

Even for powders, identifying the alloy type is quite easy, since the intensities measured for elemental concentrations higher than 0.5 % wt are accessible to any EDXRF spectrometer. Typical spectra for different alloys are shown in Figure 1. Each spectrum is automatically deconvoluted and the area under every peak is assigned as intensity for the corresponding fluorescent line. The mathematical fit assumes Gaussian shaped signals. It is also fitting the background levels.

When the energies of two characteristic lines are very close, like Ti Kβ (4.93 keV) and V Kα (4.95 keV), the advanced deconvolution algorithm is quite capable to fit the superimposed signals and calculate the individual intensities.

Figure 1: Spectral signatures of the major elements measured for different powders: (a) commercial AlSi10Mg, (b) certified IARM Ti64-18, (c) certified IARM SS 316 L and (d) certified IARM Ni-alloy 718. 

Spectra are taken with optimized conditions. Excitation of light elements requires small tube voltage, while for optimal detection of heavier elements, higher tube voltages are desired. For spectra taken in the same energy range (like in Figure 1 c-d), it may be misleading to directly compare the peak intensities. While a voltage of 50 kV was used in both cases, for Inconel 718 a smaller current must be applied on the X-ray tube (i.e., 110 microA) just to make sure the measurement remains in the linearity regime of the detector.

Figure 2: Variations in nickel concentrations (between 10.4 and 12.7 %) in commercial SS 316L powders. 

When it comes to commercial powders for AM applications, depending on alloy type, variations in concentrations of elements will be quite common, even if the powder is purchased from the same manufacturer. 

If the alloy type remains unchanged, comparison of spectra measured with fixed excitation conditions will allow screening of differences (see Figure 2). 

Figure 3: Traces of vanadium and copper found in different batches of commercial SS 316 L powders from the same supplier. Different colors represent different batches.

Figure 4: Traces of molybdenum (left image) and nickel, copper and tungsten (right image) in Ti64 powders purchased from different suppliers. Different colors represent different batches.

Certificates of analysis provided by powder manufacturers are not always complete. Elemental traces are indicated only for industry-specific requirements, as it would be the concentrations of oxygen and nitrogen for Ti64 powders [9].  While the presence of some trace elements (Figure 3) is not necessarily detrimental for the AM printed part, there are industries (like aeronautics and medical applications) with more strict requirements asking for careful screening of titanium alloys. 

Tungsten inclusions may cause cracking and a resultant reduction in yield strength and elongation [10], while Nickel is not tolerated well by all patients with a medical implant [11]. 

Only ED-spectrometers with an excellent resolution, coupled with exceptional sensitivity, would enable the user to perform XRF screening for these two elements in Ti64 alloys. In XRF spectrometry, high resolution means that fluorescent lines which are close in energy will be either sensed as individual signals or having a reasonable line overlap. 

When searching for contaminants one needs a good strategy!

The chance to detect an intruder via XRF is higher when the contaminant has one or more “signature elements” that are specific only to its composition and do not exist in the main feedstock. 

Let’s take for example the situation of the SS 316 L and Inconel 718. Composition-wise, these two alloys have some elements in common: Cr, Fe, Ni and Mo. It is only Inconel 718 that has a significant concentration of niobium (typically in the region of 4 – 5 %). If the measuring program is fine-tuned to detect also the Nb signals, then identifying the contamination of SS 316 L with Inconel 718 would be possible.

Even when the contaminant has specific “signature elements”, if their concentrations are not into % range, then, in mixtures having minute amounts of foreign powder, establishing with certitude that a specimen is contaminated may be difficult. There is no unique, sharp threshold of minimum concentration from which we can say: YES, this aliquot is contaminated. One can expect that each mixture (feedstock – contaminant) will have a typical lower limit from which contamination is proven beyond doubt. 

Compared with full solid metals, XRF tests on the powders suffer also from the randomness in particle arrangements inside the measuring cup. This aspect cannot be neglected into the general design of experiments. When feedstocks have spherical particles with a relatively narrow particle size distribution and small average diameter, particles will have the ability to arrange in relatively compact configurations. With an increase in particles’ diameter, but keeping a narrow PSD, the packing will have more voids. 

Taking all these aspects into account it is important to establish a baseline expectation for each individual fresh feedstock. Screening the composition of powder directly after atomization would allow manufacturers to better tune the process and to extend the certificate with at least a warning referring to trace elements. On the other side, when a new feedstock batch enters the AM facility, screening would alert the users to potential significant differences (in composition) compared with the previously bought batch of virgin powder. 

Figure 5: Variations in nickel concentrations for virgin IARM 718. The arrow indicates the perfect superposition of five spectra taken on a single aliquot. 

For a virgin feedstock, comparing the relative standard deviations of (i) multiple measurements on a single aliquot and (ii) a collection of single measurements on several aliquots, would be enough the estimate the variations in the XRF signals associated only with the stability of the spectrometer (i) and the influence of randomness in the arrangements of particles (ii).

Figure 6: (a-b) Examples of spectra acquired on virgin powders of Ti64 from two separate suppliers. Supplier A has a clear source of tungsten contaminating the particles. The small variation of intensities measured (a) proves that all particles contain W in their composition. Supplier B has a clean process and there is no significant peak detected on the spectral position of W Lα (8.39 keV). If the powder from this supplier is artificially contaminated with particles of pure tungsten (c, d) then the spectrometer shows the W Lα signals plus the influence of the random arrangements of the contaminant. Higher contamination levels (d) give stronger signals.

If a powder gets contaminated already at production (either because of the atomization process or because the contaminant was already in the composition of the bulk metal prior to atomization), then XRF analysis will pick up the signal of the undesired element in all measured specimens. For incidental contamination, significant variations of intensity will be observed for the “signature element” of the contaminant (Figure 6). In the specimens analyzed and illustrated in Figure 6(c) the contamination level is constant: 100 ppmv (parts per million per volume) of pure tungsten particles. Based on particle size and material density of the virgin Ti64 and pure W particles, calculating the contamination level corresponding to these mixtures puts in perspective how small this contamination is: 16 particles of W per 100,000 particles of Ti64. 

For higher concentrations of tungsten, the signals of W Lβ-lines (9.67 and 9.96 keV) will become detectable (Figure 7) making it possible to distinguish the presence of this element even in batches of Ti64 in which nickel has unexpectedly high concentrations.

Figure 7: Superposition of spectra for Ti64 batches. The tungsten-free powder containing Ni (0.21 %) and Hf (0.03 %) was kindly provided by Dr. Joseph Oluleke from Carpenter Additive, Cheshire, United Kingdom. The other three spectra are on artificial mixtures of Ti64 powders with (10000, 1000 and 100 ppmv) of pure W.

When contamination levels are small, screening of spectra should be done on a statistically relevant population of samples. In this context, the notion of “repeatability” as defined in ASTM and ISO test methods is not directly adequate for powder studies. Imagine you have a small powder amount (like 5 grams), which is doped with the addition of a single type of particulate contaminant. Visualizing this situation at microscale, this mixture is equivalent to millions of balls having one color and few balls of another color. Each time we shake the mixture and pour it in the XRF analysis cup, there is a certain probability of finding in the analyzed volume at least one ball from the contaminant. Differences in material density or particle sizes may also change these statistics. Heavier particles or smaller size contaminants have a higher chance to land at the bottom of the measuring cell, so increasing the potential to detect the signature elements of the contaminant. 

The higher the contamination level, the higher the probability of having a larger number of contaminant particles in the field of observation. To give you a flavor of the statistical variations when measuring powders, Table 1 gathers the data measured on 10 separate aliquots from each specimen of pure or contaminated powder. The contaminant is SS 316 L. 

For Ti64 contamination with SS 316L, there is a gradual and obvious increase in the average fluorescence signals specific to steel: Fe, Cr, Ni and Mo (Table 1 and Figure 8). Again, the excellent deconvolution algorithm used in the Epsilon software proves it’s strength by calculating the individual contributions of the Cr KA (5.41 keV) and V KB (5.43 keV) signals. As expected, the signals from the major elements:  Ti and V do not suffer much from the intervention of the foreign particles, but there is a noticeable increase in data spread (i.e., higher standard deviations compared to pure Ti64).

	Ti 6-4 alloy	Contaminant: SS 316 L
Element	Virgin	+ 100 ppmv	+ 1000 ppmv	+ 10000 ppmv
Ti	1131502 +/- 7273	1127706 +/- 13889	1134347 +/- 14953	1109178 +/- 10844
V (on Kβ)	2678 ± 19	2645 ± 56	2656 ± 92	2348 ± 108
Cr (on Kα)	120 ± 9	159 ± 21	522 ± 68	3965 ± 771
Cr (on Kβ)	15 ± 2	20 ± 3	67 ± 10	514 ± 100
Fe	3024 +/- 17	3235 +/- 53	5093 +/- 304	22297 +/- 3932
Ni	163 +/- 3	188 +/- 6	389 +/- 25	2227 +/- 325
Mo	242 +/- 2	357 +/- 17	1432 +/- 74	11287 +/- 873
Table 1: Statistical data, showing (averages +/- standard deviations) for measured signals in specimens of Ti64 without and with increased levels of contamination with SS 316L powder. When no indication is given, the data correspond to fluorescent Kα – lines.

Figure 8: Superposition of spectra for virgin Ti 64 (magenta) and (green) after contamination with 10000 ppmv of SS 316L powder.

Element Conc. (%)	IARM Ti64P-18	IARM Fe316LP-18	IARM Ni718P-18
Al	6.47	trace	0.49
Ti	matrix	trace	1.01
V	4.24	trace	minor
Cr	trace	17.9	19.6
Fe	0.216	matrix	17.0
Ni	trace	13.9	matrix
Nb	trace	trace	4.95
Mo	trace	2.81	3.13
Table 2: Concentrations of major elements in IARM certified reference powders

If the composition of the virgin feedstock has signature elements identical to the contaminant (as it is with the couple: Inconel 718 / SS 316L (contaminant), concentrations as in Table 2), detecting the contaminant at low levels is far from trivial.

Comparing the intensities measured for the main elements (see Table 3) in mixtures of Inconel 718 (main feedstock) / SS 316L (contaminant), none of these signals seems to show a clear trend when more contaminant is added to the mixture. For these tricky situations, we need to have a thorough understanding of advanced XRF. Calculating the expected analysis depth for each signal (data in Table 4) will lead us to choose a slightly different strategy for data interpretation.

	Nickel alloy 718	Contaminant: SS 316 L
Element	Virgin	+ 100 ppmv	+ 1000 ppmv	+ 10000 ppmv
Cr	249701 +/- 1479	245953 +/- 2424	249570 +/- 1761	248607 +/- 3175
Fe	297536 +/- 1798	292816 +/- 3128	297694 +/- 2204	299078 +/- 3947
Ni	843968 +/- 5565	829301 +/- 9494	842403 +/- 6604	835134 +/- 11915
Nb	267907 +/- 1594	267969 +/- 690	268653 +/- 506	266700 +/- 895
Mo	159000 +/- 812	158997 +/- 392	159417 +/- 350	159046 +/- 566
Nb(Lα)	31787 +/- 833	31815 +/- 670	32898 +/- 534	32607 +/- 935
Mo (Lα)	23301 +/- 800	23879 +/- 647	24871 +/- 520	24731 +/- 937
Fe/Ni	0.353 +/- 0.0002	0.353 +/- 0.0003	0.353 +/- 0.0002	0.358 +/- 0.001
Fe/Cr	1.192 +/- 0.0007	1.191 +/- 0.001	1.193 +/- 0.0007	1.203 +/- 0.002
Ni/Cr	3.380 +/- 0.004	3.372 +/- 0.006	3.375 +/- 0.003	3.359 +/- 0.006
Mo.Kα/Nb.Kα	0.593 +/- 0.0005	0.593 +/- 0.0002	0.593 +/- 0.0002	0.596 +/- 0.0005
Mo.Lα/Nb.Lα	0.733 +/- 0.006	0.751 +/- 0.008	0.756 +/- 0.008	0.758 +/- 0.009
Table 3: Statistical data, including (averages +/- standard deviations) for measured signals on several fluorescence lines (Kα and Lα) for specimens of Inconel 718 without and with increased levels of SS 316 L contamination.

	Analysis depth (µm)
Element	Kα	Kβ	Lα	Lβ1
Al	1.64	1.83	--	--
Ti	24.21	30.86	--	--
Cr	38.34	49.50	--	--
Fe	32.44	42.44	--	--
Ni	33.38	44.09	--	--
Nb	135.20	185.36	4.22	4.70
Mo	155.56	196.66	4.90	5.19
Table 4: Analysis depths when measuring with ED-XRF on a typical nickel alloy type Inconel 718 (similar composition as IARM 718-18, in Table 2).

The signals for Kα-lines of niobium and molybdenum are probing inside the mixtures up to depths of approximately 135 – 155 microns. For powders with particle sizes under 50 µm in diameter, this scale means measuring stackings of roughly 3 - 5 layers.

For the same two elements, the energies of the L-lines are small, so analysis depths are about 4 - 5 microns. These lines are more sensitive to the “surface region”, basically looking into what is in close contact with the foil supporting the sample. From an XRF perspective, the analysis of specimens with more SS 316L added into nickel 718 powders would be sensitive to a depletion of Niobium (since steel SS 316L does not have a significant concentration of this element). As a consequence, when the ratio of signals ( Mo.Lα/Nb.Lα ) exhibits a significant difference (compared to the virgin feedstock) that could be an indication of contamination.

Conclusions

EDXRF is an excellent tool for detecting foreign metal contaminants in metal powders as a relatively large quantity of powder is subjected to analysis and analysis duration is short. Although the technique does not distinguish signals originating from individual particles, it exhibits good ability to distinguish various concentrations of contamination present within the analyzed powders. 

When developing an XRF measurement strategy it is good to trace the process steps and process history to identify the potential contaminants to look for. But once the procedure is in place the whole exercise becomes almost a push-button operation. 

For severe contamination, EDXRF will be always able to indicate which are the signature elements of the contaminant, helping laboratories to improve their powder manipulation and quality management. 

Acknowledgments 

The work presented in this paper was partially sponsored by the ASTM International Additive Manufacturing Center of Excellence (AM CoE). Some of the results are included in the ASTM WK80171 “New Guide for Additive Manufacturing of Metals - Feedstock Materials-- Measurement and Classification of Feedstock Contamination”. 

Special thanks to our partners in this project: Aneta Chrostek-Mroz and Steven Hall, from MTC – Manufacturing Technology Center, Coventry, UK. They have provided the powders with controlled amounts of contaminants. 

References 

1. J. A. Slotwinski, E.J. Garboczi, P.E. Stutzman, C.F. Ferraris, S.S. Watson and M.A. Peltz, “Characterization of Metal Powders Used for Additive Manufacturing”, J. Res. Natl. Inst. Stand. Technol (2014) vol. 119, pp: 460 - 493. / Available on http://dx.doi.org/10.6028/jres.119.018
2. E. Santecchia, P. Mengucci, A. Gatto, E. Bassoli, S. Defanti and G. Barucca,  “Cross-Contamination Quantification in Powders for Additive Manufacturing: A Study on Ti-6Al-4V and Maraging Steel”, Materials 2019, 12(15), 2342.  /  https://doi.org/10.3390/ma12152342
3. J.A. Slotwinski and E.J. Garboczi, “Metrology Needs for Metal Additive Manufacturing Powders”, JOM (2015), vol. 67, pp. 538–543 / DOI: 10.1007/s11837-014-1290-7 
4. ASTM F3049 - 2014 (2021): Standard Guide for Characterizing Properties of Metal Powders Used for Additive Manufacturing Processes 
5. ISO/ASTM 52907 – 2019: Additive manufacturing — Feedstock materials — Methods to characterize metal powders
6. ASTM E539 – 2019: Standard Test Method for Analysis of Titanium Alloys by Wavelength Dispersive X-Ray Fluorescence Spectrometry 
7. ASTM E572 – 2021: Standard Test Method for Analysis of Stainless and Alloy Steels by Wavelength Dispersive X-Ray Fluorescence Spectrometry 
8. ASTM E2465 – 2019: Standard Test Method for Analysis of Ni-Base Alloys by Wavelength Dispersive X-Ray Fluorescence Spectrometry
9. ASTM E1409 - 2013 Standard Test Method for Determination of Oxygen and Nitrogen in Titanium and Titanium Alloys by Inert Gas Fusion American Society for Testing and Materials 
10. A.D. Brandao, R. Gerard, J. Gumpinger, S. Beretta, A. Makaya, L. Pambaguian and T. Ghidini, “Challenges in Additive Manufacturing of Space Parts: Powder Feedstock Cross-Contamination and Its Impact on End Products, Materials 2017, 10(5), 522. /https://doi.org/10.3390/ma10050522
11. C. A. Baumann and B. D. Crist, “Nickel allergy to orthopaedic implants: A review and case series”,  J Clin Orthop Trauma. 2020 Jul; 11(Suppl 4), pp : S596–S603