32 nd FACSS/51 st ICASS Conference Metals Determinations with High Accuracy and Traceable Uncertainty T. M. Rettberg, S. Evans Norris, J. MacIntosh VHG Labs 276 Abby Road Manchester, NH 03103 USA Quantitative analysis of materials has always involved trade-offs between high accuracy and well-defined uncertainty against and the need to compromise in order to accommodate numbers of samples, throughput demanded by multi-element determinations or lack of appropriate calibrants. In addition, ninety-nine plus percent quantitative accuracy may be rendered moot since many inorganic materials have uncertain or unstable composition, e.g. the additional weight due to waters of hydration or surface oxides. Yet, certain areas exist where the need for highly accurate measurements of well quantified uncertainty supersedes any potential compromise. These samples/industries include: Standards Certification Line in the sand applications Metals (esp. valuable precious metals and ores) New compound characterization (research) Forensics Nuclear materials For these applications, conventional instrumental analysis that feature a simple, comparative calibration followed by a series of samples may provide an inadequate result, because the impact of calibration error, instrument drift, matrix issues, and other sample measurement uncertainties can contribute to variability. The ideal goal minimal variability--can be stated as: Successive or separate observations of the system will yield the same numerical result. The use of alternating sample-standard-sample-standard... methods offer advantages and are better able to provide meaningful uncertainty values, assuming one has taken the appropriate number of individual measurements. Methods that feature an internal standard reference are even more able to track measurement drift. The National institute of Standards and Technology (NT) published a series of articles that elucidated analytical protocols designed to maximize the capability of ICP-OES spectroscopic instrumentation to deliver accurate quantification of samples with statistically well-defined and traceable uncertainty (1-3). At VHG Labs, the NT protocol approach has been applied to a large range of elements and samples for ICP-OES, as well as ICP-MS, and in this paper experiences are reported. The intent of this protocol is to provide a method for instrumental analysis that provides reliable results along with SRM traceability and statistically meaningful uncertainty. Additionally, it must be applicable to a wide range of elements and, ideally, techniques. 1. M. L. Salit, G. C. Turk, A. P. Lindstrom, T. A. Butler, C. M. Beck III, B. Norman; Anal. Chem. 2001, 73, p. 4821. 2. M. L. Salit, G. C. Turk; Anal. Chem. 1998, 70, p. 3184 3. M. L. Salit, R. D. Vocke, Jr., W. R Kelly; Anal. Chem. 2000, 72, p. 3504.
Key Features of NT protocol 1. Calibration using NT SRM and internal standard spiking The calibration relationship is one of comparing the analyte to internal standard () ratio for the test unknown to that of a well characterized standard, e.g. NT SRM materials. Since in-sample ratios are calculated, detection modes that feature simultaneous measurement of at least two channels (analyte and ) offer the greatest advantage. For the ICP-OES part of this work, the SCD detector on the Perkin-Elmer Optima 3300RL has such capability. Test Sample unknown + spike Standard (SRM) + spike SPIKE of KNOWN QUANTITY UNKNOWN QUANTITY SPIKE of KNOWN QUANTITY NT SRM, KNOWN QUANTITY 2. Alternating sample-standard sequence and repetition of the calibration block Standards and test samples are prepared in multiples (e.g. 4ea.) and measured in an alternating fashion. This manner of sequencing is able to track changes or drift in the measurement and builds an adequate measurement population for statistical analysis. Ten repeats of every measurement are taken. ID SRM 1 Test Sample 1 SRM 2 Test Sample 2 SRM 3 Test Sample 3 SRM 4 Test Sample 4 Detector Channel Action measurement #1 measurement #2 measurement #3 measurement #4 measurement #5 measurement #6 measurement #7 measurement #8 repeat block (5 times) grams 0.0018 0.0016 0.0014 0.0012 0.001 0.0008 0.0006 0.0004 0.0002 0 low weight1 All aliquots were measured gravimetrically to obtain higher accuracy and to avoid uncertainty in density both which would be introduced had volumetric measurements been used. Data (below) show that long-term weighing precision of about 1.2 mg was achieved by VHG Labs chemists (<0.05% for weighed amounts used in the protocol). Std. Dev. for 2 week/3 person weighing test low weight2 low weight3 high weight1 high weight2 high weight3 Bal#1 3-place Bal#2 3-place
3. Calculations The mass fraction of the unknown test sample was calculated according to equation 1. Calculated concentrations are traceable to NT SRM standards and are, in fact, direct daughters of SRM s. (IIR) TEST UNK x [mg ANALYTE /g SPIKE ] NT SRM Mass Fract. TEST UNKNOWN = (1) (IIR) NT SRM SOLN x [g TEST SOLN /g SPIKE ] TEST UNK where IIR refers to the Internal Intensity Ratio of signal for the analyte to its internal standard and likewise for the external standard (NT SRM). Both A-type and B-type uncertainties are tracked and processed throughout the calculations and are expressed in conventional standard deviation. The Welch-Satterthwaite formula is used to estimate degrees of freedom and coverage factors are applied (typ. k factors ~2.0) 4. Drift correction Using data for the alternating sample/ and standard/ sequence, a multi-order plot can be created and is used to normalize measurements for the entire data set. An example of a typical drift pattern is shown below. Drift Pattern 1.04 1.03 y = 4E-10x 6-4E-08x 5 + 2E-06x 4-5E-05x 3 + 0.0005x 2-0.0026x + 1.0089 1.02 1.01 1 0.99 0.98 0.97 from Traceability Tool Version 1.1 0.96 0 5 10 15 20 25 30 35 40 45 Run Order 4. MS Excel spreadsheet software tool for data processing and graphical presentation of data NT has made available a software package for calculations. Instrument data (typ. 800 signal measurements) are imported into Traceability Tool (ver. 1.1) as a grid of raw intensities or counts. These values instantaneously derive a final result, however several useful graphical displays of raw data are generated. Like the common utility of reviewing signal vs. concentration calibration plots, these graphics allow the user to visually qualify the run and/or to ascertain slight issues that can be exploited to further improve the method or sample preparation steps. Raw data appear good However, every 8 th data point starts a trend, therefore sample preparation merits investigation Raw Signal for and channels Individual ratio/mean ratio Thousands 600.00 500.00 400.00 300.00 200.00 100.00 0.00 1 5 9 1317212529333741454953576165697377 Run # Raw signal data trends 1.1 1.05 1 0.95 0.9 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 Sample # ID Ratio trends
4. measurement measurement of two signals--which are then ratioed to one another--results in a value free of much of the short term variability that may occur on account of the instrumentation used. ICP based techniques have numerous short term noise sources due to sample introduction and plasma processes. These will affect analyte and internal standard channels alike, hence will fall out of the ratio value. The example (below left) shows how effective the cancellation of short-term noise can be for ICP-OES signals and the advantage of simultaneous measurement (below right). "A" Raw counts: (top) & Int. Std.(bottom) -- Noise applied simultaneous and equal to both channels "B" Raw counts: (top) & Int. Std.(bottom)-- same level of noise as in "A", but random to both channels 300000 300000 250000 250000 200000 200000 150000 150000 100000 100000 50000 50000 0 1 4 7 101316192225283134374043464952555861646770737679 0 1 4 7 101316192225283134374043464952555861646770737679 Resulting ratio of two signals having random error, simultaneously measured Resulting ratio of two signals having random error, sequentially measured 2.400 2.500 2.200 2.000 1.800 1.600 1.400 1.200 1.000 1 4 7 10 13 16 19 22 25 28 31 34 37 40 Raw Ratio 2.000 1.500 1.000 0.500 0.000 1 4 7 10 13 16 19 22 25 28 31 34 37 40 Raw Ratio Results and Discussion Consideration of various spectroscopic conditions were first made with particular attention to ensuring that the analytical wavelengths or masses were free from spectral interferences. This was be done by evaluation of each such line or mass. Issues related to matrix or detector linearity were managed primarily by matrix matching (to a reasonable degree), as well as choosing a good signal measurement range by appropriate dilution of samples to optimal concentrations. Hence, the NT protocol was best utilized for samples of which concentration could easily be manipulated by dilution. Analytical conditions defined to meet the requirements of the method, however are not included here for reasons of brevity.
The NT protocol has been utilized at VHG Labs for over two years for a wide range of elements (see below). The primary use of the protocol has been the certification of single element standards for ICP, ICP- MS, and AA. Prior to certification as standards, these materials are prepared from carefully assayed raw materials. A reasonably accurate gravimetric value is obtained from the preparation itself, and these values were compared to that obtained with ICP-OES and the NT protocol (NP- ICPOES) calculations. The graph (below) indicates the relative agreement between these values. NP-ICPOES Results/Gravimetric conc. 1.03 1.02 1.01 1 0.99 Ag Al AuBe Bi B Te Zn Y Fe K Li Pd AgGd Hg Pr Co Eu La Pt Ho Ca CrCu LuMgMn NbNd Ni Sc Si Sn Sn W P Sb Cd Dy Er Ge In MoNa Pb Se Ta ThTi Tl Tm Yb Re Tb Rh Sb Si V Zr Sm Sr U Zr 0.98 0.97 0.96 0.95 Ag Al Au Be Bi B Ca Cd Co Cr Cu Dy Er Eu Fe Ag Gd Ge Hg Ho In K La Li Lu Mg Mn Mo Na Nb Nd Ni Pb Pd P Pr Pt Re Rh Sb Sb Sc Se Si Si Sm Sn Sn Sr Ta Tb Te Th Ti Tl Tm U V W Yb Y Zn Zr Zr The majority of samples demonstrated agreement to within +/- 0.5%. This represents about a factor of three improvement over comparable values obtained by conventional methods (i.e. those that feature external calibration followed by a several replicates of the test sample).
Below is shown a portion and example of the final-results spreadsheet page. On it one can view the relative uncertainties contributed by the measurements themselves as well as the stated uncertainty of the SRM. In the majority of cases, the stated uncertainty in the NT SRM is the primary contributor to the uncertainty budget. It is encouraging that instrument measurement error is typically under 0.05%. 0.50% 0.45% 0.40% 0.35% 0.30% 0.25% 0.20% 0.15% 0.10% 0.05% 0.00% uncertainty budgets (relative %) SRM meas. TS meas. SRM Certificate As stated previously, the goal of minimizing variability is one of having successive and separate determination of unknowns yield the same result. At VHG, inventory retain samples between 18 and 24 months of age since the original determination were re-run by NP-ICPOES. This included 64 separate samples and a wide range of elements like those shown above. The agreement of these values to their original assay was quite good with an average of 99.9% +/- 0.47% relative agreement as shown in the following table.
18 to 24 month (later) sampling: Measured/Nominal conc. 52 elements represented 1.02 1.01 1 0.99 0.98 0.97 0.96 0.95 One of the astonishing aspects of the NT protocol approach to analyte determination is that once the samples and standards are weighed and spikes of the internal standard added, the samples can be grossly mis-handled and yet still yield good results. This is due to the fact that the measurement ratio is the key factor. Whatever changes to the solutions take place will affect each element uniformly, thus not affecting the ratio. Operationally, this can be of benefit once care is taken to prepare the solution samples, the fate of the solutions is no longer of primary importance. To illustrate this, a gold (Au) sample that had been run by NP-ICPOES was subsequently and purposefully adulterated by up to +/- 5% by adding additional water, dil. HNO 3, or by evaporating a portion of the each of the eight prepared samples. The results for these samples were nearly identical and with similar uncertainty (see table below). Nominal conc. NP-ICPOES Expanded conc. uncertainty Initial value Au 1.0 1.0470 0.0022 Adulterated Samples Au 1.0 1.0468 0.0024
Alternative application of NP-ICPOES: Assay of Ni in major tungsten/iron matrix Elements in complex materials can be determined once method development has found: Appropriate wavelengths for both analyte and internal standard Characterize emission spectra for each λ and solution, choose acquisition conditions accordingly (e.g. read times) Matrix match SRM-containing standards with test samples Take ICPOES measurements and process data with traceability tool software For this analysis, Sc was used as an internal standard (424.683nm) and the analytical wavelength for Ni was at 231.6nm. The unknown had a nominal value for Ni as determined by an alternative technique. For comparison purposes, a synthetically prepared sample of similar matrix was prepared from clean materials. Below are shown the final results. Synthetically prepared sample results nominal conc. Measured Conc. Expanded uncertainty Ni 0.9536 0.9563 0.007 Unknown test sample results nominal conc. Measured Conc. Expanded uncertainty Ni 8.0 8.0012 0.0505 NT Protocol for ICP-MS determinations ICP-MS has multi-element capabilities similar to emission spectrometry. Internal standard referencing is employed routinely. Detection limits are much greater and would allow the application of the NT measurement protocol to lower concentrations. As a disadvantage most ICP-MS systems have fast, but not simultaneous, count measurement capability. Also, it would be important to have a similar ionization potential of the analyte to that of the internal standard or to have matrix-matched samples and standards.
In this experiment, Ni-containing sample at 10 ug/g was assayed using a PQ2 ICP-MS (analog detector mode). 60 Ni was determined using 55 Mn as the internal standard. Results for NP-ICPMS were compared to the prepared concentration to within 0.13% with an expanded uncertainty of under 0.50% -- both quite good for ICPMS. Conclusion While being efficient only for low throughput operations with generally well characterized matrices, the NT protocol and traceability tool software provided a superior technique to exploit the utmost capability of spectrometric instrumentation Accurate measurements can be obtained as well as well-defined uncertainty NT SRM traceability is obtained, in fact, samples are direct daughters of an SRM (when NT SRM Reference Standards are used). Software tool for data processing is rapid. Coupled with graphical viewing of data, verification of the data quality can be done quite easily.