This article discusses current and emerging tools based on tunable diode laser absorption spectroscopy (TDLAS) for greenhouse gas detection and monitoring, and concludes with some of the technical challenges to overcome in the quest to realize "ubiquitous monitoring."

As scientists, we all know that good science demands good data. In the study of global warming, obtaining good data demands widely deployed, accurate, and reliable sensors for identifying, understanding, and controlling the origins, sources, sinks, and fates of greenhouse gases, especially carbon dioxide (CO

). Sensor networks will ideally monitor and map, spatially and temporally, greenhouse gas concentrations with sufficient sensitivity and resolution to distinguish local sources from ambient background and provide fast health and safety danger alerts. These networks will be supplemented by methods for synthesizing, distributing, and using the data provided by the sensors. 

This article discusses current and emerging tools based on tunable diode laser absorption spectroscopy (TDLAS) that address these needs, and concludes with some of the technical challenges to overcome in the quest to realize "ubiquitous monitoring." 

Biogenic and anthropogenic methane sources both contribute to greenhouse gas loading. Anthropogenic sources can be controlled and limited if their origins are located. Significant sources are landfills, bovine farms, and the natural gas system, which in the United States includes nearly 500,000 active wells, 300,000 miles of transmission pipelines, and over 1,200,000 miles of distribution pipelines. Natural gas leaks are potential safety risks as well as greenhouse gas sources. Thus, maintaining the system's security and integrity is a legally regulated process of monitoring pipeline flow conditions to recognize abnormal events that may indicate leaks and ruptures, supplemented by scheduled periodic walking, driving, or aerial surveys for locating and repairing leaks.  

With increasing national emphasis on natural gas as an abundant energy resource, there is increased emphasis on improving leak detection and mitigation. Recent studies show considerable uncertainty in the leakage rate from gas production sites — it ranges from 0.42% (1) to 11% (2). Table I, extracted from a recent study prepared by the office of U.S. Senator Edward Markey, summarizes the national impact of this leakage (3).  

Table I: US unaccounted for gas, emissions, and significant incidents on natural gas systems (3)			

High leakage rates have the potential to offset the climate benefits of natural gas over other fossil fuels. Some leakage is believed to be intermittent — but significant — and thus is not detected with routine infrequent leak surveying (4). Continually operating cost-effective leak detector networks, when located judiciously, can help identify these intermittent sources. Continuous monitoring can also provide early warning of abnormal gas concentrations that may indicate a potential safety hazard.  

 is created by combusting fossil fuels, including natural gas, and is participating in climate change (5). To slow or reverse the trend of increasing atmospheric concentration of CO

) is emerging as a strategy for restraining the anthropogenic CO

 from combustion effluent, transporting it via pipeline to storage sites, and injecting it at high pressure into storage reservoirs several thousand feet deep. In essence, the process is the opposite of natural gas extraction, but uses very similar pipeline infrastructure. The current 3900 miles of CO

 pipeline in the US is projected to grow to 120,000 miles to support GCS deployment (6), and these pipelines will be subjected to leak detection and repair much like the natural gas pipeline infrastructure. Furthermore, monitoring CO

 at sequestration sites is critical for assuring that the sequestration process is safe and performing its intended purpose — that is, storing CO

 without leakage. Much like natural gas leaks, any leaks from underground sequestration infrastructure, including reservoirs, are likely to be diffusely spread through the ground to emerge from a relatively broad area and be transported by the wind. Thus, economical distributed surface sensor networks with high sensitivity and precision are needed to distinguish leaking CO

. The measurement technologies must be sensitive enough to discern CO

 leakage signals that may barely rise above the variations in natural background CO

 concentrations occurring on diurnal to interannual time scales and provide spatial resolution sufficient to localize leak origins. 

Sensors based on near-infrared wavelengths (NIR, 1.0–2.5 μm) midwave-IR (MWIR, 3–12 μm) TDLAS are emerging to fulfill these greenhouse gas sensing needs. TDLAS sensors provide fast, highly sensitive measurement of a selected gas or set of gases in complex mixtures. They couple telecommunications-style diode lasers with straightforward spectroscopic principles and detection techniques (7).  

Near-IR TDLAS emerged from the laboratory as a practical analytical tool about 20 years ago (8) and is now a proven mature technology (9,10) commercially successful in industrial markets such as process monitoring and control, quality assurance, environmental sensing, plant safety, and infrastructure security that demand sensitive and gas-specific detection of abnormal conditions with minimal false positive occurrences and down time. Indeed, some of the largest present applications of TDLAS are in the natural gas industry for pipeline leak surveying as well as monitoring pipeline gas quality. The remote methane leak detector (RMLD, Heath Consultants Inc.) is an example (11). Enabled by a state-of-the-art low-cost, low-power TDLAS platform, the detector was introduced commercially in 2005. Nearly 2500 of these handheld units are now in use worldwide for walking surveys of natural gas transmission and distribution pipeline networks. Surveying with the detector from vehicles and aircraft has been demonstrated, as has its ability to sense methane emissions from landfills. Units reconfigured as a standoff CO

 pipeline leakage and human breath (12).  

Figure 1: CH4 absorption spectra in the MWIR fingerprint region and NIR overtone, illustrating the 200Ã linestrength advantage of the fundamental.			

Attractive features of commercial NIR TDLAS sensors include reliable laser sources, high precision, very low power consumption (<1 W), compact design, minimal maintenance requirements, and an acceptable cost of ~$10,000 per sensor unit. Using sensitive detection techniques such as wavelength modulation spectroscopy (WMS), NIR TDLAS sensors probe overtones of fundamental molecular rotational-vibration energy-state transitions (Figure 1) and can sense over a dozen simple molecules at parts-per-million concentrations (Table II). The noise equivalent absorption (NEA) for NIR TDLAS sensors is typically on the order of 0.00001 at 0.1 Hz. 

Table II: Some gases measured by near-infrared TDLAS			

During the past decade, semiconductor quantum cascade lasers (QCLs) and interband cascade lasers (ICLs) have emerged as miniature industrial quality MWIR laser sources suitable for sensing in the molecular fingerprint spectral region (13,14), where absorption linestrengths are orders-of-magnitude stronger than the NIR. The MWIR region offers opportunities for sensing complex molecules that are difficult to detect with NIR TDLAS, as well as sensor miniaturization by enhancing sensitivity to simple molecules over shorter optical pathlengths. Ongoing research efforts to reduce MWIR sensor noise and power requirements by use of uncooled pulsed or low-power laser sources and small area uncooled detectors, and to reduce MWIR laser cost by high-volume production, is starting to make MWIR TDLAS sensors commercially attractive in niche markets that cannot be addressed by NIR. 

The TDLAS platform provides a technical foundation for a suite of sensors configured specifically for greenhouse gas sensing. Figure 2 illustrates the fundamental components of a TDLAS sensor: the laser transmitter, measurement path, and signal processor. Usually the laser transmitter and signal processor functions are combined in a "control unit." The laser transmitter unit generates a wavelength-agile (that is, tunable) laser beam. An optical fiber may conduct the laser beam from its origin at the laser transmitter to the measurement path. At the beginning of the measurement path, the laser beam enters a gas mixture sample. At the opposite end of the measurement path, the laser beam impinges upon a photodetector that converts the laser beam's power, attenuated by target gas absorption, into an electrical signal. The signal processor unit interprets the electrical signal and outputs information via a local display and a remote communications signal.  

Figure 2: Fundamental components of a laser-based gas analyzer.			

Some common measurement path configuration examples include extractive, point, long-path, in situ, and standoff. Each of these is relevant to greenhouse sensing. They are described and illustrated briefly here. 

Point and extractive sensors comprise an optical cell installed in a measurement chamber through which gas flows either passively or by active pumping (Figure 3). The optical pathlength is designed to provide the required sensitivity. Multipass optics, which reflect the laser beam many times between a pair of mirrors, can provide an extended optical pathlength within a small volume. Uses: Permanent or portable sampling to measure local target gas concentrations or locate leaks. 

In a long-path sensor, a transceiver projects the laser beam onto a remote surface, and receives laser light returned from the surface (Figure 4). A simple reflective highway sign is a suitable target for pathlengths of ~1000 ft. Uses: Continuous and permanent pipeline health and safety monitoring; fugitive emission localization via multisensor networks. 

Figure 4: Long-path sensor configuration.			

In situ sensors (Figure 5) have optical configurations like long-path sensors, but the laser beam transits a process gas stream (for example, a combustion effluent). Use: Monitoring and controlling CO

Figure 5: Example of an in situ sensor transceiver.			

Standoff sensors (Figure 6) are like long-path sensors but project the laser beam onto a noncooperative surface and collect a fraction of the passively scattered laser light. They deduce the amount of target gas in the laser path between the transceiver and the illuminated surface. Use: Surveying pipelines. 

Figure 6: Standoff TDLAS, exemplified by the remote methane leak detector (RMLD, Physical Sciences Inc.).			

Mobile sensors (Figure 7) are standoff sensors installed on a vehicle or aircraft (15). Uses: Mapping gathering fields, sequestration sites, and landfills; surveying pipelines. 

Figure 7: Experimental aerial standoff TDLAS on a small unmanned quadrotor. Adapted from reference 15.			

The TDLAS sensor configurations described above can be included in a spatially distributed network of sensors that communicate among each other or with a centralized network interface to enable spatial and temporal mapping of gas concentrations. The use of fiber optics enables multiplexing TDLAS sensors to monitor multiple target gases in each measurement path (16). 

Figure 8 illustrates an example of a long-path sensor network under consideration for continuous monitoring of natural gas and carbon dioxide pipelines. The eye-safe laser beams traverse across and above a pipeline to detect leaks and, upon detection, the sensor transmits alerts to the pipeline operator. The sensors ignore temporary disruptions because of passing vehicles and are insensitive to vehicle emissions. 

Figure 8: Pipeline monitor concept: (a) Example installation of open-path pipeline leak monitor. (b) Bird’s eye view of network protecting extended length of pipeline. Red arrows illustrate laser beam paths.			

In recent early testing of this configuration, remote sensor technology was adapted into permanently installed continuous and autonomous long-path sensors for CO

 pipeline leak detection and alarm reporting (17–20). Pictured in Figure 9, the transceiver, installed in a weatherproof NEMA 4 enclosure, mounted to a post with a colocated solar panel and a local RF link to a remote computer that stored data, processed the data to recognize leak signatures, and transmitted alarms via a cellular internet connection.  

Figure 9: Long-path CH4 sensor at test site.			

These sensors operated continuously for months at three field test locations: the PSI facility in Andover, Massachusetts, the Illinois Basin — Decatur Project (IBDP) GCS validation site, and the Pacific Gas and Electric training facility in Livermore, California. At each site, target gas leaks were manufactured artificially or resulted from maintenance of pipeline infrastructure. Data were acquired to demonstrate sensor efficacy, evaluate sensor response to leaks in typical operating scenarios and weather conditions, and verify sensor freedom from false alarms including any resulting from sensitivity to other ambient gases. The data reveal that CO

 leaks create distinct statistical signatures: Turbulent leak plumes cause rapid fluctuations that are readily distinguished from relatively slow natural background variations (for example, diurnal fluctuations). Figure 10 shows example data from the CH

 sensor data are similar. These wireless, solar-powered sensors, combined with proprietary statistical processing algorithms, have demonstrated the capability to discern the signatures of small leaks; detected CH

 leaks as small as 0.5 scfh, comparable to the gas flow rate of a pilot light; and activated real-time notification of leakage. When fully deployed, the envisioned pipeline protection network will facilitate identifying fugitive leaks that contribute to greenhouse gas loads and pose potential personnel safety hazards. 

Figure 10: Example long-path CH4 sensor data illustrating diurnal variation of ambient CH4 (upper plot); period of leakage from a source upwind of the laser path (lower plot).			

For widespread deployment of low-cost point sensor networks, prototype chip-scale low-power integrated-optic gas-phase sensors based on MWIR TDLAS are in early development. Figure 11 illustrates a sensor element and TDLAS package concept intended for low-cost mass production and low power consumption (21). Practical challenges to fabricating these sensors include selecting and designing the high-Q micro-resonator sensing element appropriate for the selected analyte; coupling laser light into and out of the sensing element; and device thermal management, especially stabilizing laser temperature with the precision needed for sensitive spectroscopic detection. Novel gas sensing elements using low-loss silicon resonant waveguides, providing effective optical pathlengths that are orders of magnitude greater than their physical dimensions, will permit compact integration of a laser source, sampling elements, and detector in configurations suitable for inexpensive mass production. Recently developed interband cascade lasers that operate at room temperature with low power consumption are expected to serve as MWIR sources (22). Proven grating couplers efficiently transfer laser light into and out of the sensing elements. The package capitalizes on the use of simplified low-cost laser sources having no supplementary optics or thermoelectric controller.  

Articles

Columns: Lasers and Optics Interface

In the study of global warming, obtaining good data demands widely deployed, accurate, and reliable sensors for identifying and understanding the origins, sources, and fates of greenhouse gases. Sensing technologies based on tunable diode laser absorption spectroscopy are starting to provide some of this critical data.

Current and Emerging Laser Sensors for Greenhouse Gas Detection and Monitoring

Laser-induced breakdown spectroscopy (LIBS) is a technique that works well for a wide range of applications. In this interview, Dr. Richard R. Hark, a professor in the Department of Chemistry at Juniata College in Huntingdon, Pennsylvania, discusses his work with LIBS in applications such as forensic science, conflict minerals, and geochemical fingerprinting. Part II of this interview will focus on Hark's work using LIBS for emergency response to hazardous materials along with an interview with a first-responder who has been involved in that research. 

Early in your career you synthesized a large number of novel ninhydrin analogs as reagents for visualizing latent fingerprints on porous surfaces. Did that research naturally lead to your interest in laser-induced breakdown spectroscopy (LIBS)?

My background in forensic science did lead directly to my involvement with LIBS. I became familiar with the technique in 2002 and immediately saw the advantages of applying LIBS to the analysis of trace evidence. 

I came to Raman spectroscopy in a more roundabout manner. Since 1999 I have taught a class called "The Chemistry of Art," which is an interdisciplinary course that explores the intersection of chemistry with the visual arts. In this laboratory-based class we learn about artists' materials, issues facing conservation scientists, and many basic chemistry concepts as we explore the chemistry and history of art media such as paints, dyes, metals, alloys, ceramics, glass, plastics, paper and fibers, and photographic materials. Because of the desire I had to get involved in research related to cultural heritage objects, I was able to spend a sabbatical leave at University College London and the Victoria and Albert Museum in 2007–2008. While there, I was privileged to work on a number of projects involving the analysis of medieval manuscripts and miniatures using Raman spectroscopy. 

How have you incorporated LIBS into your undergraduate teaching and research efforts?

Our undergraduates have been using the technique in course work and research since I acquired our first LIBS instrument in 2003. LIBS has been a topic in our sophomore analytical chemistry course as well as the subject of upper-level special topics courses over the past decade. More than 20 undergraduate students have been involved in LIBS research projects over the past 11 years, resulting in numerous conference presentations and several papers. Last fall, Juniata was the site for the first Laser-induced Breakdown Spectroscopy for Undergraduate Research and Teaching (LIBS-URT) conference. The second LIBS-URT will be held again at Juniata October 10–11, 2014. Our goal is to promote awareness of the technique among faculty who teach and do research with undergraduate students. 

LIBS can be used for multiple applications in forensic science, including glass, paint, ink and paper, counterfeit currency, ammunition and gunshot residue, fingerprints, wood, fibers, and biological materials. Are there other materials encountered in forensic studies that might be suitable for analysis by LIBS? What are the limitations for its use in forensic science?

In addition to the types of evidence mentioned, LIBS has application for the analysis of drugs and their synthetic precursors, hair, soil, explosives, and nuclear material. Every analytical technique suffers from disadvantages that must be understood and taken into account so that the method can be appropriately applied to forensic evidence. While there are some inherent limitations of the LIBS technique, principally related to matrix effects, the limit of detection, and level of precision, the fact that LIBS has been utilized in very few court cases probably represents the greatest hurdle to more general use of the method. The admissibility in court of a particular type of scientific evidence must be established using the 

 standards (1,2) before a new analytical technology gains widespread acceptance. It also takes time for forensic practitioners to develop familiarity with, and then confidence in, new methodology. 

Has LIBS reached its full capacity in forensic science yet?

Since LIBS offers the ability to carry out rapid, minimally destructive elemental analysis directly on virtually every type of material of importance to criminal investigations without the need for sample preparation, I believe that LIBS has great potential in the field of forensic science. As more portable, ruggedized LIBS instruments are developed and made commercially available, the use of the technique as a high-throughput screening tool at crime scenes and to collect data of evidentiary value will likely increase. This is an evolving field that is moving from an initial phase, in which preliminary studies demonstrated a proof-of-concept, toward a more mature stage where best practices and standard operating procedures (SOPs) are being developed, known rates of false positives and false negatives are being established, and LIBS is gaining greater visibility among forensic scientists. 

You have also used LIBS to analyze paint samples, artworks, and archeological artifacts. What challenges did you encounter in using LIBS for that research? What is the most interesting result you've found so far in that work? 

I used LIBS on some Neolithic skull fragments from Jordan as part of a project to learn if the discoloration of bones was because of natural causes or the application of an iron oxide pigment, but the results were inconclusive. We have not utilized LIBS on oil paintings or other more sensitive artworks, but the technique would find good application with analysis of frescoes. For my work with cultural heritage objects I tend to use only nondestructive techniques such as Raman and X-ray fluorescence (XRF). We did carry out a preliminary study that compared LIBS to scanning electron microscopy–energy-dispersive X-ray spectroscopy for depth profiling of automobile paint. We created cross-sectional graphs that reflected the ablation of successive layers of the paint finish by plotting the LIBS emission line intensities versus the number of laser pulses. 

Your recent research has involved analyzing conflict minerals using LIBS to determine their origin. What are conflict minerals and what spurred your interest in this research area?

So-called "conflict minerals" are natural resources that are mined in areas associated with armed conflict and human rights abuses and include materials such as "blood diamonds" or columbite-tantalite (known as 

). While working on a project to analyze pegmatite minerals, we recognized that our work had an obvious application to the conflict mineral problem. Because the LIBS emission spectrum provides a unique chemical signature of a material that can be used to discriminate geological specimens originating in one place from samples of the same kind from other locations, we felt that it would be possible to use this technique to verify if a particular sample came from a source controlled by a legitimate mining organization or a militant group. 

How does this fit into the broader use of LIBS in geochemical fingerprinting?

The concept of geochemical fingerprinting is based upon the fact that the Earth's crust is compositionally heterogeneous, both horizontally and vertically, and that minerals forming within the crust will reflect that inherent geographic heterogeneity. A variety of analytical techniques can be used to obtain information about the elemental composition and isotopic ratios of a geological sample and thereby establish its chemical signature. Comparison of these chemical fingerprints with a library of samples with known provenance allows an unknown sample's source to be identified. Because LIBS is a straightforward atomic emission spectroscopic technique that delivers rapid, multielement detection in real time with minimal sample preparation, it is being increasingly applied to the analysis and provenance verification of geomaterials. 

How does LIBS fit into the overall set of analytical methods that can be used for geochemical fingerprinting? What are its strengths and weaknesses compared to other methods?

Many different analytical methods, such as X-ray diffraction (XRD) and XRF spectrometry, electron microprobe analysis (EMP), instrumental neutron activation analysis (INAA), and inductively-coupled plasma analysis (ICP, ICP coupled to mass spectrometry [ICP-MS] and laser ablation with ICP-MS [LA-ICP-MS]) have been used within the Earth, environmental, and archaeological science communities for geochemical fingerprinting. In addition to the general benefits that have already been mentioned, LIBS specifically offers several important advantages that make it a useful analytical technique for geochemical materials, including the fact that LIBS is particularly sensitive to light elements (such as H, Li, Be, B, and C), that it provides high spatial resolution allowing for stratigraphic analysis and mapping of complex samples, and that it can be combined with complementary spectroscopic techniques such as Raman or laser-induced fluorescence (LIF). While the typical limits of detection (low parts-per-million) and the level of precision (5–10% RSD) for LIBS experiments are normally not as good as some of the methods used for analysis of geomaterials, they are often sufficient to provide discrimination between samples of different provenance. 

What impact do you expect handheld LIBS to have on geochemical fingerprinting?

The possibility of using portable or handheld LIBS instruments in the field is perhaps the most attractive feature of this technique that has yet to be exploited, not only for geochemical fingerprinting but also for many other applications. If the cost of a handheld LIBS unit were comparable to or ideally less than that of a handheld XRF instrument, I would foresee many opportunities for such a device to have a significant impact on many fields. Compared to XRF, and assuming that a handheld unit had similar analytical figures of merit as a laboratory model would, LIBS could offer a lower level of detection, the ability to see all elements, and higher throughput than handheld XRF. 

What are the next steps in your research on geochemical fingerprinting?

To more fully establish the benefits of the LIBS technique for geochemical fingerprinting, we are analyzing large sample suites of volcanic rocks and carbonate minerals with known, unambiguous provenance. The creation of spectral libraries with a substantial number of samples allows us to use robust chemometric analysis tools for verification that samples of unknown origin can be reliably matched to their geographic point of origin. This is also the next step in our work with the conflict minerals project, but it has proven challenging to obtain samples from a sufficient number of mining locations in central Africa. We are also continuing our work with obsidian studies as this has important archeological applications such as the determination of the provenance of obsidian artifacts. 

What other types of research can be advanced by using LIBS to study mineral samples?

The recent development of laser ablation molecular isotopic spectrometry (LAMIS) holds the potential for LIBS to be used to measure isotope ratios in geomaterials. LAMIS is based on the fact that isotopic shifts associated with molecular spectroscopy are substantially larger than those in atomic spectra. Ultrahigh resolution spectrometers are therefore not required to measure isotopic ratios in minerals for which the ablation process creates diatomic molecular species such as metal bromides, chlorides, fluorides, iodides, and oxides. The possibility of using LIBS to quantitate some radiogenic nuclides used in geochronology would be particularly exciting. 

You are applying LIBS to a diverse set of analytical questions. For those applications, is LIBS still at a stage of having to prove its value compared to other analytical methods? In what other areas do you think LIBS is likely to make an important contribution? Are there any applications for which LIBS is, or is likely to become, the primary, "go-to" analytical method?

The LIBS technique is still undergoing a maturation process for many of the research areas in which I am involved, but steady progress is being made. In some industrial applications, such as the in situ analysis of molten metals and recyclable materials, LIBS has already established itself as the ideal method. It is important to recognize that although LIBS has many advantages, it is not the best solution for every application. The key is to implement the method in situations where the cost of the instrumentation is reasonable, the sensitivity and selectivity of the LIBS measurement are sufficient, and the versatility, portability, stand-off capability, and robustness of the approach are crucial. The use of a LIBS unit on the Mars rover Curiosity demonstrates that in some cases LIBS is definitely a mission-critical technique. 

Part II of this two-part interview will focus on Hark's work on using LIBS for emergency response to hazardous materials. We will also feature a second interview on this area of LIBS research, with Adam L. Miller, of John R. Plumer Associates, LLC, who has worked as a first-responder. Miller and Hark have worked together on the use of handheld LIBS for first responders. Part II will be published in our November 2014 issue.  

Daubert v. Merrell Dow Pharmaceuticals, Inc.

Spectroscopy Spotlight

Dr. Richard R. Hark, a professor in the Department of Chemistry at Juniata College in Huntingdon, Pennsylvania, discusses his work with LIBS in applications such as forensic science, conflict minerals, and geochemical fingerprinting.

Laser-Induced Breakdown Spectroscopy: A Closer Look at the Capabilities of LIBS

This installment of "Atomic Perspectives" is the second in a series describing the educational components and processes necessary in teaching and learning the technique of X-ray fluorescence (XRF) spectroscopy. Although three of the four states of nature (gas, liquid, and solid) are discussed briefly, the focus is on the sample preparation of bulk solid materials. Because of the advances in XRF instrumentation hardware and software as well as algorithm knowledge and selection for mathematical corrections, selection of an appropriate sample preparation technique is the major source of error in the most exacting quantitative analyses. The discussion begins with the fundamental assumptions necessary to prepare a bulk powder material for the best possible accuracy during quantitative analysis. Analysis of error is considered to show where sample preparation fits in the scheme of things. Then we discuss the physics involved during the excitation and emission processes and consider the major sources of error in preparing powders. Finally, the basics of grinding, pressing, and the fusion method of sample preparation are described. 

The analytical process involved in producing an X-ray fluorescence (XRF) spectrochemical analysis of bulk materials (such as samples from ore deposits, ship cargo holds, quarries, coal seams, rail cars, manufacturing processes, and production lines) can be described as following four main categories: representative sampling and subsampling from the bulk, taking an aliquot from the subsample from which a specimen for analysis is prepared, generating net intensities from the specimen in an XRF spectrometer, and converting the net intensities into concentrations through a calibration curve. As shown in Figure 1 (1), the XRF technique can analyze most any material. The question is how much effort if any should go into the preparation of the specimen (the portion that is actually analyzed). As described in the previous installment of this column series on XRF spectroscopy (2), the reason to determine the elemental composition at high concentration of many materials with an XRF spectrometer is that the technique has the capability to produce the highest accuracy and precision possible. To achieve that goal, it is necessary when perfecting the method to review the sources of error and the amount that each may contribute (equation 1). 

Equation 1, the analysis of variance, shows contributions of the various components making up the analysis system. Analysis of variance (3) shows that the total error is equal to the square root of the sum of the squares of error of the individual components of the error; that is, the error from the counting statistics, the power source, the spectrometer, the macro sampling process, the specimen preparation process, the reference standards used for calibration, the type of correction algorithms applied and the decisions made when making those corrections, and everything else. Everything else can include the temperature, humidity, vibration, and dust in the laboratory and surroundings. These components contribute both random and systematic error. It is the job of the analyst performing the method development to minimize random errors and remove the systematic errors, which are generally quite large and easy to remove by an experienced spectroscopist. In fact, because of the advances made in the stability of the generators, tubes, and electronics, the accuracy of the goniometers, and the empirical and theoretical correction algorithms available in the software, the largest error now comes from standard selection, sampling, and specimen preparation. Therefore, the choice of method and application of technique when performing the specimen preparation is of paramount importance to the final accuracy of the analytical method. Notice the word accuracy is used here and not precision. In fact, it is possible when using XRF to choose a method where the precision will be very high but the accuracy will not be acceptable. The differentiation between precision and accuracy needs to be well understood when working with XRF, and will be discussed below. 

	If the goal is to achieve the highest accuracy possible, it is absolutely necessary to understand the concepts involved. For sure, the XRF technique for quantitative analysis is a comparative technique. That is, by virtue of using calibration curves, the ratio technique, or type standard technique, the closer the standards used are to the mineralogy, particle homogeneity, particle size, and matrix characteristics of the unknown, the more accurate the analysis will be. Let us call this "The Golden Rule for Accuracy in XRF Analysis." Conversely, the farther standards and unknowns are from each other in the above listed characteristics, the less accurate the analysis will be. This is a certainty.

 can be defined as the closeness or agreement among replicate results. 

 can be defined as the nearness of a result to the true or accepted value. The bull's-eye on the left in Figure 2 shows that several determinations can lead to high precision (the dots are close together), but poor accuracy (they missed the bull's-eye). This can often be the case with the pressed powder method of specimen preparation. The bull's-eye on the right shows both high precision (dots are close together), and high accuracy (they hit the bull's-eye). This is almost always the case with the fusion method of specimen preparation. Later on, we will discuss why this is so, and how to choose which method will give the fastest and most accurate result. It is also important to understand that, as stated before, XRF spectrometers produce intensities, which are converted into concentrations through a calibration curve. The intensities themselves can be of very high precision, and the theoretical precision can be calculated because it is equal to the square root of the total number of counts. It is therefore possible to determine if the spectrometer is performing properly by making a 10-shot precision test and comparing it to the theoretical precision of each element. This is highly recommended before preparing standards or unknowns, making intensity measurements, and performing the calibrations. 

Basic Concepts for XRF Sample Preparation

The general rules for XRF analysis apply to situations where the highest accuracy is desired; these rules include

		The specimen should be representative of the bulk sample

		The specimen must be homogenous and flat

		Standards and unknowns must be very similar (density, particle size, particle homogeneity, mass attenuation coefficient)

		Specimen preparation should have good reproducibility

		The specimen should remain stable before and during analysis

More specific considerations involving the physical properties of the specimen (and container, if liquids or loose powders are analyzed) include

		Surface film absorption characteristics (liquids)

		Particle size effects (loose and pressed powders) such as grain size effect, inter-mineral effect, and mineralogical effect

		Grinding curve analysis (pressed powders)

		Groove depth, smearing, and groove orientation (metals)

	As an ideal starting point, the specimen should have an analytical depth for each wavelength (energy) analyzed that provides 

infinite thickness. That is, the thickness of the sample will not affect the analysis. On a more practical note, the specimen should have an effective layer thickness that will provide at least 99% of the analysis. Figure 3 shows the concepts of infinite thickness and effective layer thickness. The wavelengths of interest are not excited in the white layer. They are excited in the light blue layer, but most of them are absorbed by the sample matrix. Wavelengths in the dark-blue layer are excited and escape from the specimen to take place further in the analytical process. Modern software has the capability to calculate the effective layer thickness, and such a calculation is shown in Table I. It is worthwhile to take note of the narrow depth (thickness) from which most of the analysis is taking place for each element. For instance, sodium is coming from a depth of only 4 Âµm, aluminum and silicon from around 10 Âµm, and all of it from only 200 Âµm. The average thickness of a human hair is about 100 Âµm, so it is evident that the surface of the prepared specimen cannot be touched, breathed on, or exposed to any conditions that could alter it. Moreover, the effective layer thickness for any given wavelength changes with the matrix. As an example, the effective layer thickness for iron K-alpha wavelength changes from 3000 Âµm (3 mm) in carbon to 11 Âµm in lead. Togel (4) gives the equation for calculating the effective layer thickness; the equation of calculation and geometry of the situation are shown in Figure 4. The problem of particle size effect is shown in Figure 5, where the specimen analyzed is neither homogenous nor representative in the effective layer analyzed for the wavelength of interest. This sample must be crushed and ground to obtain a representative, homogenous, effective layer thickness to be analyzed for the wavelength of interest shown (and for any longer wavelengths). Even then, the mineralogical effect may cause problems, which cannot be solved by any means other than fusion. We have seen first-hand the problems caused by the mineralogical effect when analyzing kyanite, sillimanite, and andalusite (three polymorphs of Al

). In their pure form, each has the same weight percent of silica and alumina; however, using one as a standard to analyze the other can result in analysis totals ranging from 75% to 125%! Figure 6 illustrates the mineralogical effect on iron intensities from minerals prepared as pressed briquettes having the same particle size and concentration of iron. Because the compositions of mineral deposits vary from one location to the next, in addition to substituting various elements as contaminants in the crystal structures, it is virtually impossible to closely match the matrix of standards to unknowns using the pressed powder technique; therefore highly accurate analyses at high concentration levels are difficult if not impossible for many natural materials. 

Gases, Liquids, Loose Powders, and Metals

The analysis of gases is not practical with an XRF spectrometer, although it is possible. To be sure, there are many other analytical techniques better suited to this task. The specimen preparation for liquids and loose powders is a matter of choosing the appropriate equipment and accessories, and it is better demonstrated in a workshop environment such as at the annual XRF course at ICDD in Newtown Square, Pennsylvania (5), or at the Denver X-ray Conference Workshop on XRF Sample Preparation (6). Metals are also a matter of trial and error, with the properties of hardness and smearing weighing heavily on the choice of cutting tool, sanding device, or sanding material (7). In any case, the surface should be renewed whenever it has changed, and that depends on the element and the material it is in. Selecting the sample spinner on the spectrometer will remove the problem of groove orientation affecting intensities. And in any case, the same fundamental concepts of infinite thickness and effective layer thickness (described above) apply no matter the physical state of the material.

Now that the question of "How long should I grind?" has been answered, we can take a closer look at the grinding and pressing (also known as 

) process. There are many different types of grinding equipment, and the size of the specimen and briquette have to be taken into consideration when choosing one, as well as the standard sizes and characteristics of the manufacturer of the XRF instrument being used. The typical size of a briquette made for a typical XRF spectrometer ranges from 30 to 40 mm in diameter, and involves about 5 g of sample and 1 g of binder. Figure 7 shows a briquette (pressed pellet) of cement. The powder was poured into an aluminum cap to provide stability to the pellet. Also shown are the ring, puck, and container in which the specimen was ground. The weight and motion of the ring and puck mill are quite efficient for reducing most oxide materials down to about 30â40 Âµm particle size in about 3â4 min. Grinding further will lead to agglomeration of the particles, thus defeating the purpose of the grinding exercise. Adding a binder is necessary for the stability of the briquette, depending on the friability of the material. Anzelmo and colleagues (8) describe adding a grinding aid that greatly facilitates the uniformity of the particle size reduction, eliminates contamination from the grinding media and from the sample sticking to the grinding media, speeds up the cleaning process, and reduces dust during the cleaning process. Manual or automatic presses operating between 20,000 lb and 40,000 lb have been found adequate as long as the time of the pressing procedure remains constant. After the grinding equipment, additives, and backing equipment are decided on, the final step is to determine the optimum grinding time, given the goal of achieving the correct effective layer thickness for all wavelengths of interest. Figure 8 shows plots of grinding time versus intensity for various briquettes. The intensities are changing with different grinding times, so in addition to finding the time to achieve the appropriate effective layer thickness for each wavelength of interest, it is desirable to find a time when intensities are not changing drastically with time. 

When should one use the pressed powder technique? When parameters such as high speed of analysis, low skill level for preparation, small concentration range of calibration, and lowest limit of detection are involved, the pressed powder technique may serve the purpose adequately. The technique can be especially useful when working with man-made products, because the material being analyzed is well known and characterized when the process is working correctly; secondary standards can be collected and analyzed by wet and alternative methods to provide "matrix matched" standards, allowing the prospect of a highly accurate analysis. A problem arises with the technique when the process is not working correctly, and the unknowns are then not matrix matched. This presents a challenging if not unacceptable situation to a process control situation requiring accurate analyses.

The borate fusion technique shown in Figure 9 consists of mixing a ground sample with a borate flux (lithium or sodium) inside a 95% platinum, 5% gold crucible, heating to about 1000 Â°C with agitation until the flux melts, and then dissolving the sample homogeneously in the flux (10). The choice of lithium borate or sodium borate depends on the application, but lithium borates are more commonly used because of the greater retention of moisture on the surface of the glass disks with sodium borate flux, and the desire to analyze for sodium. Lithium borates exist as lithium tetraborate (Li

). It is not necessary to melt the sample, so temperatures above 1100 Â°C are not necessary and can adversely affect the retention of volatiles and flux. A nonwetting agent (bromide or iodide) can be added when integrated into the flux, or independently, depending on the application. All oxides dissolve in borate flux, so it is essential to make sure the sample is oxidized. Metals alloy with the platinum forming a eutectic, which lowers the melting point of the platinum and results in its destruction. If the sample is not oxidized fully, it may be necessary to add an oxidizer to the sampleâflux mixture, and heat at a lower temperature to start, and then raise the temperature after the oxidation process has taken place. Careful observation of the oxidation process will allow commonly analyzed volatiles such as sulfur (9,10) and even chlorine to be analyzed. After the sample is completely dissolved in the flux, the casting process can be performed by pouring the melt into a heated 95% platinum, 5% gold mold. This particular alloy of platinum is chosen to prevent wetting (sticking) of the bead to the mold. The cast glass disk is cooled quickly at first to freeze the bead in the amorphous (glass) state, and then it is cooled faster over a period of time with forced air to relieve the stresses in the glass disk from the initial drop in temperature when first cast and to allow the bead to be retrieved by hand for analysis. The entire process is shown in Figure 10. Glass disks made from different materials and cast into molds of different sizes are shown in Figure 11 along with their crucibles. 

The process can be performed manually with a Meker burner and a ring stand or with a good muffle furnace. Automatic fusion machines, electric or gas (propane or natural gas), can produce multiple glass disks automatically depending on how many positions they have (typically one to six). For the highest sample throughput, it is possible to use a robot to automate the entire process, including the weighing and flux dispensing step. 

	The advantages of fusion are listed below:

		Provides excellent homogeneity and flatness

		Creates perfectly matrix matched standards and unknowns

		Allows for matrix correction when previously not possible

		Can use with the synthetic standards for calibrations, when reference standards are otherwise not available

Easy to produce "defendable" analyses (traceable to International Primary Reference Standards)

		Can provide wide-range (universal) calibrations

		Should be used whenever it is not known if the standards and unknowns are not matrix matched. This is often the case when working with materials from nature (minerals mining, clays, ores, geological, and soils). It is also used in the case where man-made products are mixed with other man-made products and also products from nature such as blended cements, catalysts, electronic materials, and high-tech composite materials.

	Fusion should be used whenever it is not known if the standards and unknowns are matrix matched. This is often the case when working with materials from nature, such as mining samples, minerals, clays, ores, geological, soils, wastes, and environmental samples. It is also used in the case where man-made products are mixed with other man-made products and also with products from nature such as blended cements, polymers with additives, catalysts, electronic materials, and high-tech composite materials. 

These samples are difficult if not impossible to matrix-match with standards. An everyday example of this problem is the analysis of SiO

 in raw mix used in the manufacturing of cement (9,10). The plot on the left in Figure 12 shows the calibration curve made for SiO

 in raw mix from three cement plants (denoted by the triangle, circle, and x symbols). The scatter around the line is unacceptable and cannot even be made into straight lines by separating out each plant. Mathematical interelement corrections will not improve this situation because the problem is not from absorption-enhancement effects; it is from the mineralogical effect and can only be solved by fusion. The middle plot shows that the points have now moved very close to the line after fusion, giving acceptable accuracy expectations when analyzing unknowns. The plot on the right shows even more improvement after mathematical corrections are applied. Mathematical corrections are now possible because the larger, dominating error from the mineralogical effect has been removed, allowing the smaller error from absorption enhancement to be reduced (and to be able to see the improvement).

	The choice of specimen preparation for XRF analysis is critical to achieving the accuracy desirable. The choice between pressed 

powder briquettes and fused glass disks is actually easy to make when one knows whether the samples are matrix-matched or not. If the standards and unknowns are not known with certainty to be matrix-matched, then fusion is the safe choice to ensure that a ridiculous analysis is not reported. For more detail about the concepts and practices of these techniques, we recommend attending the training course and workshop noted in references 5 and 6.

Atomic Perspectives

X-ray Fluorescence Spectroscopy, Part II: Sample Preparation

Determination of trace metals in oil and petroleum products is typically based on inductively coupled plasma–optical emission spectroscopy (ICP-OES), which requires significant sample preparation and expensive instrumentation. This method, based on the use of a handheld energy-dispersive X-ray fluorescence analyzer, involves minimal sample preparation, uses authentic standards for calibration, gives low parts-per-million detection limits, and provides significant time and cost savings.  

Metal contaminants in combustion engines can come from a number of sources, including oil, oil additives, and engine wear. In some cases, metals are intentionally added to oil and fuels to improve antioxidant, anticorrosive, dispersing, and anti-wear properties (1). During the refining process, metal contaminants decrease catalyst activity and selectivity, alter the product distribution (2,3), and cause auto-oxidation and decomposition of hyperoxides (4,5). In engines, metal contaminants can cause corrosion of parts such as bearings, valves, and pistons (1,6), eventually leading to engine failure (2). Early warning and appropriate corrective action are essential to improve engine performance, longevity, and cost savings (6,7). Moreover, the presence of toxic metals in automotive fuels and oils represents a potential source of environmental contamination (7–12).  

As shown in Table I, a number of standard methods have been developed specifically for measuring trace metals in fuel, oil, and petroleum products. The most common methods are based on atomic absorption spectrophotometry (AAS) and inductively coupled plasma–optical emission spectroscopy (ICP-OES). These methods require significant sample preparation, which involves ashing the sample in a platinum crucible for 2–12 h, mixing with flux and heating to form a fusion, acid digestion to dissolve the residues, and dilution to a known volume (12,13). This process is labor intensive, time consuming, and expensive. In addition, the process can lead to erroneous results because of volatilization of metals or inadvertent introduction of metal contaminants from reagents and sample containers (1). After the samples are prepared, they are analyzed via AAS or ICP-OES. This process involves introduction of the sample into a nebulizer as an aqueous solution, an oil–water emulsion, a microemulsion (1), or a three-component emulsion (14,15). The use of nonaqueous solvents affects nebulization efficiency, atomization efficiency, precision, and accuracy (16), and the industry consensus is that the use of organic solvents gives erroneously low results (1). A related technique called rotrode-OES, which is based on the use of a rotating graphite disk-shaped electrode to concentrate large wear and contaminant particles and detection of specific metals via OES (17), is documented in ASTM method 6595 (18) and is widely used by the US military and postal service as a qualitative screen for metals in unfiltered oils.  

Table I: Standard methods used to measure trace metals in petroleum products (adapted from reference 5)			

Energy dispersive X-ray fluorescence (EDXRF) and wavelength dispersive X-ray fluorescence (WDXRF) spectroscopy represent an alternative to the standard AAS and ICP-OES methods, and have been used to determine a variety of elements in petroleum cokes (11), petroleum oils (5,11,19,20), shale oil (21), lubricating oil (5), and residual oil (13). EDXRF systems can be purchased as benchtop instruments, portable units, or handheld analyzers. Numerous vendors have developed marketing literature based on the use of a benchtop EDXRF instrument or custom EDXRF instrument to measure trace metals in oil. It should be noted that some of this work is dated, uses nonportable laboratory-based instrumentation or obsolete products, and derives quantitative results through fundamental parameter based models (versus calibrating instrument response using authentic standards).  

More recently, inductively coupled plasma–mass spectrometry (ICP-MS) (22–27) and total reflection X-ray fluorescence (TXRF) (4–6) have been used for this application. TXRF is a specialized configuration of EDXRF in which a small amount of sample is dissolved or suspended in a liquid matrix, applied to a quartz disk, dried, and analyzed as a thin film at a very low grazing angle. The use of a drying step and a low grazing angle provides detection limits that are two to three orders of magnitude lower than EDXRF methods. However, the use of small sample volumes and the various preparation steps can lead to nonrepresentative and erroneous results, respectively. 

Recent developments in handheld EDXRF analyzers merit their reconsideration for elemental analysis applications (28,29). Compared to AAS, ICP-OES, and TXRF, handheld EDXRF requires much simpler sample preparation procedures and provides the convenience of handheld analysis for off-site or in-field work. The goal of this study was to develop and characterize a new handheld EDXRF method for this application based on the use of authentic standards to provide accurate quantification of V, Cr, Fe, Ni, and Zn. This article provides details for the method, analytical figures of merit (linearity, detection limits, precision, accuracy, robustness), and a comparison of results from the analysis of a set of oil samples using both handheld EDXRF and ICP-OES methods.  

A total of 22 samples previously analyzed via ICP-OES were acquired from the San Francisco Branch of Inspectorate America Corporation. These samples represent a variety of matrices, including crude oil, gas oil, diesel, hydrocracked gas oil, and bunker fuels. Samples were stored in 8-oz clear glass bottles at ambient temperature.  

Accurate quantification via XRF frequently requires the use of authentic standards to calibrate instrument response. Preparation of standards containing several different metals in an oil-based matrix is nontrivial and time consuming. Here, a commercially available stock multielement solution containing 500 ppm of several metals in an oil matrix (Conostan oil analysis custom blend multielement standard S-21) and metal-free oil diluent (Conostan element blank) were used to prepare a set of calibration standards (5, 10, 25, 50, and 100 ppm) via gravimetric dilution of the 500 ppm standard with the metal-free oil diluent.  

Samples and standards were placed into XRF sample cells, sealed with 3.5-μm Mylar (polyethylene terephthalate) film, and mixed by hand for 1 min before analysis. Analysis was performed using an Olympus-Innov-X Delta premium model handheld EDXRF analyzer equipped with a rhodium tube, several different filters, and a silicon drift detector (SDD). Although this analyzer can be operated in handheld mode, for this work it was placed in a test stand to ensure minimal user exposure to the relatively low energy (<50 kV) and low power (50 W) X-ray tube source, and controlled from an IBM-compatible portable PC using Innov-X PC software to facilitate data acquisition and interpretation. All spectra were acquired with the analyzer set to "soil" mode using "beam 2" excitation conditions (35 kV and use of a custom filter) to give the best detection limits for the target metals of interest. All sample spectra were acquired using 2-min analysis times. Three replicate spectra were acquired for the blank, standards, and some of the samples to assess reproducibility.  

Table II: KÎ± and KÎ²  energies (keV) for the five target metals			

The target metals and their principal emission energies (K

) are provided in Table II. XRF theory and nomenclature is described in more detail elsewhere (30), but in brief K refers to the shell from which the electron was removed and the subscript refers to the shell from which an electron transitions to fill this hole (α is next higher shell and β is second higher shell). EDXRF has lower resolution than WDXRF and users need to be wary of spectral overlaps. As indicated in Table II, the K

 fluorescence from V (5.43 keV) overlaps the K

 fluorescence from Cr (5.41 keV). These overlaps can lead to erroneous results, depending on the types and levels of elements in the sample (30).  

Figure 1: Full (top) and zoomed in (bottom) XRF spectra of a blank oil sample and an oil standard containing 25 ppm of several metals detectable via XRF.			

Sample spectra were downloaded from the handheld EDXRF analyzer, uploaded into Microsoft Excel, and converted into the proper format (intensity in units of counts per second versus energy in units of keV). Calibration and calculation of sample concentrations was achieved using "first principles" as described below, but it should be noted that this process can be made easier by "recalibrating" the analyzer (that is, using a calibration curve that plots the analyzer-reported concentration versus the known concentration to determine an appropriate slope and offset). Peak intensities were determined by computing the maximum intensity within ±0.1 keV of the reference K

 energies of each metal shown in Table II. These intensities were normalized to that of the maximum Compton peak intensity (in the range of 20.2–20.9 keV). This procedure, commonly referred to as 

, is used to correct for variations in sample density, matrices, and orientation (30). Results from analysis of the blank and standards were used to generate calibration curves and compute metal concentrations in the samples.  

Figure 2: Compton normalized calibration curves for five target metals. R2 values for V, Cr, Fe, Ni, and Zn are 0.998, 0.999, 0.998, 0.998, and 0.998, respectively.			

Representative EDXRF spectra of oil samples containing several trace metals are shown in Figure 1. The top spectrum shows a Compton peak at 20–21 keV and a broad bremsstrahlung peak in the range of 12–36 keV from the X-ray tube source. The bottom "zoomed-in" spectrum shows K

 peaks from V, Cr, Mn, Fe, Co, Ni, Cu, and Zn. The presence of Ni and Zn peaks in the blank are artifacts caused by their presence in the analyzer optical system and detector.  

Table III: LODs for the five target metals			

Most quantitative methods involve the preparation of a fresh set of calibration standards just before analysis. Given the limited availability and high cost of the stock solution, a single set of standards was prepared and used to calibrate the response of the handheld EDXRF analyzer throughout the 47-day course of this study. Representative calibration curves for various metals are shown in Figure 2. The use of a single set of excitation conditions gave different sensitivities for each metal, with the best sensitivity (largest slope) for Zn and slightly reduced sensitivity (smaller slopes) for metals with smaller atomic numbers. Over a one month data collection period, calibration curves consistently gave correlation coefficients (

) values of at least 0.998 for the metals of interest, demonstrating that this method provides a good correlation between Compton normalized peak intensities and concentrations.  

Figure 3: Plot of analyzer response for a 100 ppm standard as a function of time. The trends and small slopes demonstrate the stability of EDXRF response and standard concentrations over the duration of the study (47 days).			

Limits of detection (LODs) were computed using a method similar to that employed in various Environmental Protection Agency (EPA) methods (that is, using a concentration that provides a signal-to-noise ratio of 3). Three replicate spectra of the 0, 5, and 10 ppm standards were analyzed to compute the signal-to-noise ratio of the K

 peaks for each metal. The signal was computed as the difference between the maximum Compton-normalized intensity of the K

 peak of the metal of interest within ±0.1 keV of the reference line energy of the standard and the Compton-normalized intensity of the blank at the same energy. The noise was computed from the standard deviation of the Compton-normalized intensity of the standard. Because the signal-to-noise ratios are proportional to concentration, these data can be used along with the equation above to compute the LODs, which are provided in Table III. The lowest LODs were obtained for the metal with the highest atomic number (Zn), which is consistent with the slopes in the calibration curves shown in Figure 2.  

Figure 4: Plot of analyzer response for the 25 ppm standard stored at ambient conditions (AMB) and at 10 Â°C (COLD) as a function of time. The trends and small slopes demonstrate that EDXRF response and standard concentrations were relatively independent of temperature. %RSDs for V response at ambient and cold conditions were 2% and 4%, respectively; %RSD for Ni response at ambient and cold were 2% and 6%, respectively.			

Precision, as measured by the percent relative standard deviation (%RSD) of the Compton-normalized intensity, was in the range of 2–5% for standards ranging from 5 to 100 ppm. To further evaluate reproducibility and robustness of the method, XRF response was studied as a function of time, temperature, and laboratory analyst. Figure 3 shows a plot of the Compton-normalized intensity for a 100 ppm standard over a 47-day time period. Figure 4 shows a plot of analyzer response for a 25 ppm V standard analyzed at ambient and refrigerator temperature. These data, particularly the small %RSD and lack of significant trends in response (slopes < 10

) as a function of both time and storage temperature, demonstrate the stability of both XRF response and the standard concentration. Figure 5 shows a comparison of Ni concentration from two laboratory analysts in the four samples where it was above the LOD. These data, which correspond to differences of 2–5%, demonstrate that similar results can be achieved by different operators.  

Figure 5: Results from analyses of Ni in four different samples by two analysts, demonstrating reproducibility between different users.			

Results from analysis of the 22 samples were compared to those from ICP-OES, both with respect to the presence or absence of a metal and computed concentrations. For the purposes of this discussion, ICP-OES results are considered to represent the "true" values, but it should be reiterated that this method may give erroneous results as described earlier. With respect to determining the presence or absence of a metal, EDXRF results can be grouped into four categories: correct and incorrect detection (true and false positives) and correct and incorrect nondetection (true and false negatives). EDXRF correctly identified the presence and absence of the five target metals in 72% and 91% of the samples, respectively. False positives and negatives occurred in 28% and 9% of the measurements, respectively. Although these rates are rather high, it should be noted that each false positive and false negative occurred when the metal concentration was below the limit of quantification (LOQ, defined as the concentration giving a signal to noise of 10) and very close to the limit of detection. A comparison of the quantitative results is shown in Figure 6, which shows EDXRF and ICP-OES results from the determination of Ni in six samples where Ni levels were above the LOQ.  

Figure 6: Comparison of ICP-OES and EDXRF results for determination of Ni in six samples containing Ni levels above the XRF LOQ.			

Some discussion on these differences as well as the larger issue of the suitability of XRF for quantitative analysis is appropriate here. The differences between the EDXRF and ICP-OES results shown in Figure 6 were consistent between two operators analyzing these same samples, and can be attributed to determinate errors in the ICP-OES or EDXRF methods. The analysis of nonhomogeneous samples containing large particles or flakes of metals via XRF may yield erroneously high results because of the settling of these more dense materials to the bottom of an XRF sample cup. To get around this problem, one might use a single method (such as ICP-OES) to attempt to determine total metal concentrations in a sample or two or more methods to determine metal concentrations in filtered oil samples (for example, XRF or ICP-OES) and the composition of metal particulates (for example, rotrode-OES or scanning electron microscopy with energy dispersive X-ray spectroscopy [SEM-EDX]). Differences between the matrices and densities of the standards (prepared in an oil matrix) and samples (that is, crude oil, gas oil, diesel, hydrocracked gas oil, and bunker fuels in this study) can also cause determinate errors. This problem can be resolved using the method of standard additions, but it is often impractical to perform this type of calibration for each different sample. Despite these caveats, handheld EDXRF can be used to screen for the presence of metals in oil, monitor trends in metal concentration in oil over time, or accurate quantitative analysis (assuming that the samples are relatively homogeneous and have matrices that are similar to the standards). From a more practical point of view, once the calibration curves have been derived from authentic standards, the slope and offset can be programmed into the XRF analyzer's "soil" mode of operation so that the analyzer will directly compute and display the concentrations of various elements in the sample.  

Table IV: Typical times associated with determination of trace metals via ICP-OES and XRF methods			

To the best of our knowledge, this is the first publication in a peer-reviewed journal describing the use of authentic standards and a handheld EDXRF analyzer for determination of trace metals in oil. Relevant figures of merit include minimal sample preparation, multielement analysis, linear calibration curves, LODs in the range of 1–5 ppm, and robustness (that is, relative independence of results as a function of time, sample temperature, and operator). Although this method can also be implemented on a benchtop EDXRF system, handheld units have detection limits that are in many cases within an order of magnitude of benchtop systems and are more than adequate for monitoring low parts-per-million levels of metals in oil. Moreover, handheld systems are typically less expensive, much simpler and easier to operate, and can be used in field work. The two most important advantages of this method versus more widely accepted AAS, ICP-OES, and ICP-MS methods are its far simpler sample preparation procedures and faster analysis times. These advantages are illustrated in Table IV, which shows the approximate times required for analysis of a blank, five standards, and one sample, and Table V, which shows a more detailed comparison of various figures of merit for AAS, ICP-OES, EDXRF, and TXRF methods. Given its clear advantages with respect to analysis times, capital equipment costs, and per sample costs, handheld EDXRF should be considered as a viable and potentially better method for this application.  

Table V: Figures of merit for AAS, ICP-OES, EDXRF (including handheld, portable, and benchtop units), WDXRF, and TXRF methods for determination of trace metals in oil and petroleum products. The LODs for AAS, ICP-OES, and TXRF are based on preparation of oil samples, which involves use of small sample masses and dilution.			

 (1) R.Q. Aucélio, R.M. deSouza, R.C. deCampos, N. Miekeley, and C.L. daSilveira, 

 (4) K.M. Bichinho, G.P. Pires, F.C. Stedile, J.H. DosSantos, and C.R. Wolf, 

 (5) A. Cinosi, N.Andriollo, G.Pepponi, and D. Monticelli, 

 (6) N. Ojeda, E.D. Greaves, J. Alvarado, and L. Sajo-Bohus, 

 (7) M.F. Gazulla, M. Orduña, S. Vicente, and M. Rodrigo, 

 (8) M. Korn, D.S. Santos, B. Welz, M.G. Vale, A.P. Teixeira, D. Lima, and S.L. Ferreira, 

 (9) R.M. De Souza, A.L. Saraceno, C.L DaSilveira, and R.Q. Aucélio, 

 (10) C. Hardaway, J. Sneddon, and J.N. Beck, 

 (11) H. Kubo, R. Bernthal, and T.R. Wildeman, 

 (12) M.F. Gazulla, M. Rodrigo, S. Vicente, and M. Orduña,

 (13) R.M. DeSouza, A.L. Meliande, C.L. DaSilveira, and R.Q. Aucélio, 

 (14) A.V. Zmozinski, A. DeJesus, M.G. Vale, and M.M. Silva, 

 (15) A. DeJesus, A.V. Zmozinski, J.A. Barbara, M.G. Vale, and M.M. Silva, 

Peer-Reviewed Article

Determination of trace metals in oil and petroleum typically is based on ICP-OES methods, requiring signficant sample preparation and expensive instrumentation. This article presents an alternative method, based on the use of a handheld energy-dispersive X-ray fluorescence analyzer, that involves minimal sample preparation, uses authentic standards for calibration, gives low parts-per-million detection limits, and provides significant time and cost savings.

Rapid Determination of Trace Metals in Oil Using Handheld X-Ray Fluorescence Spectroscopy

Issue PDF

Click the title above to open the Spectroscopy July 2014 regular issue, Vol 29 No 7, in an interactive PDF format.

Spectroscopy-07-01-2014

Current and Emerging Laser Sensors for Greenhouse Gas Detection and Monitoring

Laser-Induced Breakdown Spectroscopy: A Closer Look at the Capabilities of LIBS

X-ray Fluorescence Spectroscopy, Part II: Sample Preparation

Rapid Determination of Trace Metals in Oil Using Handheld X-Ray Fluorescence Spectroscopy

Vol 29 No 7 Spectroscopy July 2014 Regular Issue PDF