Atomic Absorption with Computer-Controlled Sampling Walter G . Boyle and William Sunderland Lawrence Radiation Laboratory, Unicersity of California, P. 0. Box 808, Livermore, Calif. 94550
A sampling turntable and an atomic absorption spectrometer are operated by a small digital computer. Under computer control, standard and unknown solutions are randomly sampled to minimize errors due to spectrometer drift. The concentrations of the unknowns are calculated from computer-fitted calibration curves. This sampling cycle is repeated and new values for a running mean and running standard deviation are obtained. This paper shows that the precision of the analyses i s improved by using repetitive sampling cycles, electronic integration, and base-line corrections. Copper, nickel, and zinc, the major components in NBS 157a, and titanium in titanium-tungsten alloys are analyzed with a precision of *0.2% and 11.5%, respectively. The effect of additives on the determination of trace platinum (1 ppm) i s also evaluated
.
Although atomic absorption spectrometry (AAS) has become a very important analytical technique, there are many problems associated with its use ( I ) . One of the more important problems is instrumental drift. The spectrometer readings change slowly with time in a random manner, primarily because of variation in the characteristics of the hollow-cathode light source and the instability of the nebulizer-burner system. One of the ways this problem of drift can be overcome is t o run a set of standards with every set of unknowns. Moreover, this calibration and calculation cycle can be repeated a number of times and the unknown solution values that are obtained on each cycle or “run” averaged. If the sampling during each run is random, drifting values should have a tendency t o be averaged. To implement such a procedure requires computational aids. The use of computers in AAS has been mostly confined to large general-purpose computers with batch processing of the data (2-5). This approach can be most helpful for the analysis of routine samples and for long-range research. However, the turnaround time for batch processing large amounts of data is too long for a system using multiple runs and averaging. The use of small dedicated computers as a data acquisition and/or control system has been demonstrated in several recent experiments (6-8). A small on-line computer that would control a sampling device and directly process data from a n atomic absorption instrument would offer a number of advantages. It would be a useful tool t o investigate procedures because it could give running outputs, so that the operator has current knowledge of the status of the measurements. Cal(1) L. L. Lewis, ANAL.CHEM.,40 (12), 28A (1968). (2) J. Ramirez-Mufioz, J . L. Malakoff, and C. P. Aime, A d . Chim. Acta, 36, 328 (1966). (3) J. L. Malakoff, J. Ramirez-MuAoz, and W. G. Scott, ibid., 42, 515 (1968).
(4) J. L. Malakoff, J. Ramirez-MuAoz, and C. P. Aime. ibid., 43, 37 (1968). (5) R. H . Wendt, At. Absorp. Newslett., 7, 28 (1968). (6) G. Lauer, R. Abel, and F. C. Anson, ANAL.CHEM.,39, 765 (1967).
(7) S. P. Perone, J. E. Harrar, F. B. Stephens, and R. E. Anderson, ibid., 40, 899 (1968). (8) J. W. Frazer, G. D. Jones, R. Lim, M. C. Waggoner, and L. B. Rogers, ibid., 41, 1485 (1969).
ibration curves could be displayed and signal-to-noise ratios improved using electronic integration techniques (9). Finally, this arrangement would simplify sample handling and be rapid enough to employ the multiple-run averaging technique described above. This paper describes such a computercontrolled sampling system with examples of its use in atomic absorption. EXPERIMENTAL
Atomic Absorption Instrumentation. A modified JarrellAsh Model 82-536 single-beam atomic absorption instrument was used for this investigation. The instrument had the following modifications. An improved power supply for the hollow-cathode sources was designed and constructed at this laboratory. Light from the hollow-cathode source and light from a reference source were mechanically chopped at 87 Hz. The mV reference signal was provided by a Clairex photocell, a n 18-V tungsten lamp, and a variableload resistor powered by a Heath Model ERW-17 lowvoltage power supply. The two signals were fed into a Princeton Applied Research (PAR) Model 120 lock-in amplifier. A J. Fluke high-voltage Model 143C power supply was used t o supply the photomultiplier. The following three different slot-type Techtron burners were used: Model AB-40 (a 5-cm stainless steel burner for nitrous oxideacetylene), Model AB 50 (a 6-cm titanium burner designed for nitrous oxide-acetylene but used with air-acetylene because the flame was somewhat more stable than the usual IO-cm burner), and Model AB 41 (a 10-cm stainless steel burner for use with air-acetylene). A grating blazed at 3000 A, R-I36 photomultiplier (Hamamatsu T. V. Company), and a 100-p entrance slit and a 150-p exit slit were used for all the experiments. A Leeds and Northrup model W millivolt recorder was occasionally employed. Digital Hardware and Auxiliary Equipment. A digital Equipment Corporation (D. E. C.) PDP-8/S digital computer was used in this work. It has a 12-bit, 4096-word core memory and a cycle time of approximately 8.0 psec. This computer has many peripheral devices, and it is mounted on casters so that it may be moved about to service a variety of analytical instruments (10). The following input/output devices were necessary for the experiments: an ASR-33 teletype, a Hewlett-Packard 141A storage oscilloscope, a D.E.C. high-speed tape reader, three computer-controlled telephone type relays (Automatic Electric Co.), a VIDAR 260 voltage-to-frequency converter, a D.E.C. crystal-controlled timer (clock) and D.E.C. binary counter (IO). The clock and counter programming arrangement is shown in Figure 1 , where it can be seen that double-precision counting (24-bit accuracy) was available. The output signal from the PAR amplifier was fed to the V-F converter. The pulses from the converter were counted by the computer over a n accurate time interval and processed or stored. The double-precision 24-bit counter has a limit of about 8.3 X 106 counts. Overflow is signaled by the presence of a minus sign in the output. One has certain limits and options as t o integration time, full-scale voltage, and the VIDAR (9) R. K. Skogerbee in “Flame Emission and Atomic Absorption Spectrometry,” J. A. Dean and T. C. Rains, Ed., Marcel Dekker, New York and London, 1969, Chapter 13. (10) R. E. Anderson and G. D. Jones, unpublished work, Lawrence
Radiation Laboratory, Livermore, Calif.
ANALYTICAL CHEMISTRY, VOL. 42, NO. 12, OCTOBER 1970
1403
Manuo I Operations COUNTER
12 bits in memory
12 bits
Preliminary Input
i n hordwore
L O A D SAMPLE WHEEL STANDARDS FROM 1-N INCREASING ORDER
r-l
LSH = Leost significant h a l f
LOAD A N D START CLOCK
MSH =Most significont h a l f
E N T E R TOTAL
SET INTEGRATION TIME (PAUSE TIME) (BASELINE REPEAT TIME)
SET INSTRUMENT PARAMETERS
I
STANDARDS
I
/
U N K N O W NS
1
ENTER CONCENTRATION OF EACH INCREASING
COVER SLIT
u + l ~
1 iYES
UNCOVER SLIT
Figure 1. Flowchart of clock-counter functioning sensitivity setting. The time constant on the PAR amplifier (which allows electronic pre-filtering) is adjustable from 1 msec to 30 sec. The PAR amplifier has a full-scale voltage output of 10 V. This output was shunted to a value of 20 mV. No noise problems were encountered when the recorder and the computer were both in operation from a simple parallel connection. Sample Turntable. Many different designs of sampling mechanisms can be computer operated. Our sample changer uses a Technicon AutoAnalyzer turntable having positions for 40 9-ml polystyrene sample cups. The turntable is rotated by a synchronous motor controlled by a relay in conjunction with a holding circuit. A time delay relay operates a second synchronous motor to raise or lower a sampling tube. A third relay controls the direction of rotation of the turntable. These relays are, in turn, controlled from a set of relays in the PDP-8/S interface. The operation of the turntable has been very reliable and needed only an initial adjustment of the sampling tube. The turntable requires about 1 sec to change from one position to the next adjacent position. Digital Software. The programs for data acquisition and processing and for operating the sample turntable were written in assembly language. The programs were punched on cards and assembled on a PDP-1 computer using an assembly program called PALMAC-8 (11). The card input is written in a field-free format similar to PAL I11 and MACRO 8 (12, 13). The output produces a high-speed _____
~~
~~
~
t--/
I NTE GRATE CAR: C E ,l:’ NT ABSORPTION)
DARK AS CURRENT TOTAL COUNTS
I
Figure 2. Flowchart of manual operations and preliminary input printer listing with some diagnostics and a punched binary paper type that can be loaded with the D. E. C. high-speed tape reader. The D. E. C . Floating Point Package 111 with extended functions was used to do most of the arithmetic and to compute logarithms (Digital 8-5-C-S) (14). This package uses one 12-bit word for the exponent and two 12-bit words for the mantissa. Subtracting one bit for the sign gives 23 bits of significance. The random number generator was obtained from the DECUS library (15). It was slightly modified so that only two digit numbers were produced. The clock and voltage-to-frequency analog-to-digital scheme was adapted from a FOCAL patch program (16). Four assembly-language computer programs have been written for use with this system. In general, the computer programs are written so that the standards and samples may be run in a random manner. The programs operate the sample turntable in such a way as always to select the shortest path to the next desired sampling point. As sampling proceeds, the optical absorbances obtained are printed and stored until the entire set of standards and unknowns has been run.
~
~~
(11) K. Bertran, R. Anderson, and R. Bystroff, “A PAL 111, MACRO 8 Assembler for the PDP-8 to be Run on the PDP-I,” Informal Publication, Lawrence Radiation Laboratory, Livermore, Calif., 1968. (12) Digital Equipment Corp., Maynard, Mass., PAL 111 Symbolic Assembler Programming Manual, Digital 8-3-S, 1965. (13) Digital Equipment Corp., Maynard, Mass., MACRO-8 Programming Manual, DEC-08-CMAA-D, 1965. 1404
0
(14) Digital Equipment Corp., Maynard, Mass., Floating Point System Programming Manual, DEC 8 - 5 3 , 1965. (15) P. T. Brady, DECUS Program Library, No. 5-25, A Pseudo Random Number Generator for the PDP-5 Computer, Digital Equipment Corp., Maynard, Mass., 1965. (16) R. E. Anderson, “FOCAL with A/D Patch 11,” Informal Publication, Lawrence Radiation Laboratory, Livermore, Calif., 1968.
ANALYTICAL CHEMISTRY, VOL. 42, NO, 12, OCTOBER 1970
A first-order least-squares fit of the standardizing data results in an equation that is used to calculate the unknown concentrations. These concentrations are printed first and then stored in a running mean and running standard deviation. The entire set of samples and standards may be run as many times as desired. After each cycle the new values for the unknown concentrations and also (after the first run) a new running mean and running standard deviation are calculated and printed out. One of the programs (CONTROL-8) allows complete control of the sample wheel by the teletype; a second (AUTO8) runs the sample wheel automatically and randomly; a third program (BRACKET-8) automatically brackets each unknown [or set of unknowns) between two standards (still randomly); and the fourth program (AVABS) measures absorbances only, storing these in a running mean with the associated statistics. Some of these programs have an oscilloscope display of the calibrating points; Le., optical absorbance PS. concentration of the standards. The running mean is calculated according to the equation NRM = ORM.(&) n - 1
-I- NV 7
where NRM = new running mean, ORM = old running mean, NV = latest measured result, n = total number of runs including the latest run, and the running variance is calculated according to the equations NRV
=
+ (NRM -- NV)'
ORV.("+)
for n degrees of freedom, and TRV
=
NRVs(5)
for n - 1 degrees of freedom, where NRV = new running __ TRV = true variance, ORV = old running variance, and Of course, SD = d T R V , where SD running variance, is the standard deviation. Solutions and Reagents. Standard solutions were prepared from metals having a purity better than 99.9% except for nickel, which was 9 9 . 8 z pure. All other reagents were ACS reagent grade. Operation. Solutions are placed in the 9-ml plastic cups provided for the Technicon AutoAnalyzer sampling turntable. Standard solutions are placed o n the turntable first, starting with the position to be defined as standard number one, and then in increasing order according to concentration. Unknowns are placed on adjacent positions in any order starting in the first position .after the last standard. Position zero (actually position 40) is used as the blank that defines the 0% absorption line. The blank or zero position is initialized as the present position of the sampler when the program is loaded into the computer. The electronic integration time is nominally 5 sec; the pause time is 7 sec (pause time is defined as the time interval between the start of aspirating the solution and the start of the integration). The base line is read at the beginning of a run and after every fourth measurement. These parameters can be changed using the switch register. Figure 2 shows the preliminary information required by the computer. The dark current is read immediately after the last standard concentration is entered, and the entrance slit must be covered before this last entry. After the dark-current count is printed, the slit is uncovered and the first base line read. The programs proceed automatically under computer control from then on (except CONTROL-8), reading standards and unknowns randomly until all have been read, printing the information, and then repeating the procedure in a (usually) different sequence,. One complete analysis is referred to as a "run." The programs will proceed with this repetitive cycle until the computer is stopped by the operator.
Table I. Instrumental Precision of Atomic Absorption Systema Nominal 95 absorption, Absorbance, Re1 std confidence No. of mean dev, limits, runs 0.214 0.129 13 94 1.348 0.103 0.055 16 84 0.8221 0.038 16 60 0.4041 0.072 0.063 16 50 0.3104 0.118 0.209 0.111 16 20 0.1536 4 VIDAR set at 24 mV, 0.1-sec time constant, Cu hollow-cathode lamp, 3247 A.
z
z
~~
z
~
~~
RESULTS AND DISCUSSION
T o evaluate the usefulness of these programs in atomic absorption spectrometry, it was necessary to determine the instrumental limit on the precision of the modified Jarrell-Ash spectrometer. This was done by measuring the optical absorbance of neutral density filters at 3247 A (Cu radiation). The neutral density filters were placed in a firmly positioned holder with no flame in the light path. Using the AVABG program (without the sample wheel), a base line was read without the filter. The filter was then placed in position and the optical absorbance measured and printed out. This was done 13 to 16 times. A base line was read each time. Several measurements of the i d e r s were made to find an optimum setting of ;nstrument variables. Taking a large signal from the amplifier and using a high setting on the VIDAR (500 mV) gove comparable or slightly lower precisions than did the final 20-mV setting, even though the total number of counts for a 5-sec integration was about a factor of three higher; i.e., about 6 X lo6us. 2 X lo6counts at the final setting used. The 20-mV setting, a 0.1-sec PAR time constant, and a 5-sec electronic integration time resulting in about 2 X lo6 counts for a full-scale measurement gave the best precision. Longer integration times did not improve the precision. The results listed in Table I show excellent precision. Indeed, variations in these measurements might possibly be due to the necessary removal of the filter for the baseline reading, which results in slight variations in its position within the holder when replaced. The best precision was found at a n absorbance of 0.4, which agrees with Smith and Feldman (17). The value of repetitive runs can best be seen in the 95 confidence limit of the mean. This limit is about one-half the standard deviation after 16 determinations. The depression of the optical absorption of platinum by certain elements has been studied by Strasheim and Wessels (18) and by Schnepfe and Grimaldi (19). It was found (19) that an addition of 0.5 % each of cadmium and copper as the sulfate was effective in eliminating much of this interference. A study of the effect of some additives on the sensitivity of the platinum determination was made using the AVABS program to give an average absorbance calibration curve and inspect its linearity. The results are shown in Figure 3. The curvatures shown in Figure 3 could easily be seen on the scope display, By making several runs using this program even small differences in sensitivity can be seen to be real. One can quickly determine that lanthanum gives the highest sensitivity (17) S. B. Smith, Jr., and F. J. Feldman, Abstracts 8th National Meeting of the Society for Applied Spectroscopy, Anaheim, Calif., Oct. 1969, No. 143. (18) A. Strasheim and G. J. Wessels, Appl. Spectrosc., 17, 65 (1963). (19) M. M. Schnepfe and F. S. Grimaldi, Talanra, 16, 591 (1969).
ANALYTICAL CHEMISTRY, VOL. 42, NO. 12, OCTOBER 1970
1405
Absorbance, mean 0.0024
z z
Std dev 0.00030
0,00027 a Air-C2H2, 10-cm slot burner, Pt * VIDAR at 24-mV full scale. VIDAR at 48-mV full scale. 0.0024
Table 11. Determination of Trace Platinum (Repeatabilityp 95 Approx confidence total counts Re1 std dev, limits, full scale 12.4 6.2 8 X lo6 11.1 5.9 6 X lo6 hollow-cathode lamp at 2659.5 A.
and copper the lowest, even though the differences are small, and that HCl generally lowers sensitivity. Table I1 shows the reproducibility of measuring a n absorbance of 0.0024 obtained for 1.095 ppm of Pt in 0.5Z La. A 20-sec integration time was used. Since the double precision 24-bit counter has a capacity of about 8.3 X lo6 counts, to integrate for a longer time, the sensitivity of the VIDAR had to be decreased. Thus, the 40-sec integration gave fewer counts than the 20-sec integration and there was no appreciable difference in this measurement. This absorbance suggests that 1 ppm of platinum can be determined to +0.06 ppm. The determination of copper, nickel, and zinc in NBS 157a shows the accuracy and precision that can be expected from averaging many runs and bracketing the unknowns with standards. In each case three standards were prepared and the BRACKET-8 program was used to find the two standards closest to the unknown. Standard solutions that would give absorbances of about 0.4 were used. These solutions also contained about the same amount of matrix ions as the sample. The results are shown in Table 111. Again it can be seen that the 95 Z confidence limit is a good indication of the precision that can be expected using a multiple-run system. Here &0.2 is easily attained. The analyses seem t o be biased slightly high. This could be due to trace metal contamination, which may result from adjusting the nitric acid concentration. When using AAS for major component analysis, especially for the more common ions, this can be a very serious problem (1).
10
1 c
0.5% 0% HCI; 0.25% La, 0% H C I --- 0.25% Lo, 2% HCI 0.63% Cd, 2% HCI ----- 0.63% CU, 2% HCI ; 0.5% Cu + La,
---
0.5% Cd, 10% H C I
..........., 10% HCI
Platinum
readings
20 secb 40 secc
16
No. of 18
Before the analysis of 157a was attempted, several determinations of copper were made to determine the optimum base-line repeat, integration time, and PAR time constant for this well-behaved system. Some of these determinations are shown in Table IV. From this Table it was determined that reading the base line more often than after every fourth measurement did not improve the precision. A 0.1-sec time constant was optimum. These runs on copper were made many days apart, and several parameters were varied. The overall precision of the mean value for copper (treating each multiple result as a single determination) should be noted. The determination of titanium in titanium-tungsten alloys is a good illustration of major component analysis in a noisy system. This determination is done with a nitrous oxideacetylene flame, which has more flame emission noise than the air-acetylene flame. Moreover, carbon buildup in the slot of the burner increases the noise and causes drift (20). Thus, the randomization of the sampling procedure is important because of the changing absorption signal with time. The alloy is best handled by dissolving in dilute HF with the dropwise addition of nitric acid. The titanium-tungsten composition could vary over a wide range, so it was necessary to investigate the effect of both HF and tungsten on titanium absorption. Several solutions were prepared containing from 1 to 20% HF and 0 to 1000 ppm of tungsten. The absorbances were run several times using the AVABS program. The results are shown in Table V. The absorption of titanium increases with increasing tungsten concentration out to about 500 ppm of tungsten and then decreases. The decrease of titanium absorption at that point and the apparent increase in the absorption with increasing HF concentration may be due t o changes in the viscosity of the solutions. Eight titanium solutions were prepared containing 500 ppm of W and 5 % HF ranging from 0 to 150 ppm of Ti. These were run as standards using the AVABS program. A photograph of the oscilloscope picture of these absorbances is shown in Figure 4. A slight curvature in the trace is noticeable, and one point (125 ppm Ti) can be seen to be in error. Three of these solutions were run as unknowns using the remainder as standards. Each unknown was bracketed by a set of standards. The programs AUTO-8 and BRACKET-8 were used and the results compared. Table VI shows that the use of the bracketing program resulted in a significant increase in the precision. This is because when bracketing is used so that only one unknown is bracketed by two standards, only four measurements are necessary for each complete analysis. When the complete calibration curve is used, eight measurements are needed, thus allowing the system more time to drift. The differences in accuracy occur because the cal-
- m/ml
Figure 3. Analytical curve for Pt with various additives 1406
Integration time
(20) P. E. Thomas and M. D. Amos, Resonance Lines, 1, 1 (1969).
ANALYTICAL CHEMISTRY, VOL. 42, NO. 12, OCTOBER 1970
Table III. Analysis of NBS NBS analysis, wt %
Wt % foundb (mean)
Re1 std dev, %
limits, %
58.61
58.60 58.75 58.76
0.65 0.36 0.30
0.35 0.19 0.17
20 sec
16 16 15
Zinc
29.09
29.15 29.19
0.38 0.41
0.21 0.22
5 sec 10 sec
15 16
Nickel
11.82
11.91 11.90 11.89
0.57 0.37 0.46
0.31 0.19 0.25
5 sec 10 sec 10 sec
16 18 16
Element Copper
a
ation time 5 sec 10 sec
No. of runs
Air-CzH,, 6-cm slot burner (AB-SO), VIDAR at 24-mVfull scale, bas: line read every fourth solution, 7-sec pause, PAR lime constant 0.1
sec. Hollow-cathode lamps were Cu, 3247 A; Zn, 2139 A; Ni, 2320 A.
Unknown solution concentrations were Cu, 11.7 ppm; Zn, 3.79 ppm; Ni, 17.9 ppm. Table 1V. Copper Determinations on NBS 157aa 95 %
cu," ppm, mean
Re1 std dev, %
confidence limits, %
Base line repealc
5.882 5.853 5.886 5.877 5.878 5.879 5.870 5.897 5.876
0.56 0.43 0.57 0.37 0.50 0.49 0.69 0.85 0.95
0.30 0.24 0.31 0.22 0.39 0.38 0.53 0.65 0.73
1-1 1-4 1-2 1-4 1-4 1-4 1-4 1-1 1-1
Time constant, sec 0.1 0.1 0.1 0.1 0.1 0.03
- ---
Pause lime, sec
Integration time, sec
No. of runs
12 7 7 7 7 7
5 5 5 5 5 5
16 15 15 13 9 9
-
-
Solution of NBS 157a lo give 5.866 ppm Cu. Average Cu: 5.878 =t 0.012 ppm. Cu bollow-cathode lamp at 3: = 1+ 4 means the base line was read after every fourth measuremet b
-
-
Table V. Variatio1I in Absorbance for 100 ppm of Ti in the Presence of W and HF. w 11,
HF, %
200
0
5 10 20
nnm
~I(... 500
1000
n ,7QC ".I,""
" 171c ".LILY
" ,
