Anal. Chern. 1981, 53, 1655-1658 1655

(6) Ekimoff, D.; Mathews, S. E.; Walters, J. P., Unpublished work. (7) Walters, J. P. Appl. Specfrosc. 1972, 26, 323-354. (8) Ekimoff, D. Ph.D. Thesis, University of Wisconsin-Madison, 1981, (9) Walters, J. P. I n "Contemporary Topics in Analytical and Clinical

Chemistry"; Hercules, D. M., et ai., Eds.; Plenum Press: New York,

(10) Olesik, J., unpubllshed work. (11) Mathews, S. E.; Niernczyk, T. M.; Walters, J. P. Appl. Spectrosc.

(12) Kaiser, H.; Wairaff, A. Ann. Phys. (Leipzlg) 1939, 34, 297. (13) Hosch, J. W.; Walters, J. P. Appl. Opt. 1977, 16, 473-482. (14) Klueppel, R. J. Ph.D. Thesis, University of Wlsconsin-Madison, 1979. (15) Brewer, S. W.; Walters. J. P. Anal. Chem. 1969, 41, 1980. (16) Takahashl, T. Bunko Kenkyo 1966, 15, 164. (17) Washburn, D. N., unpublished work. (18) Scheeline, A.; Coleman, D. M.; Wakers, J. 1'. Appl. Spectrosc. 1978,

32, 215-223. (19) Hurd, P. A. "Metallic Materlais: An Introduction to Metallurgy"; Hoit,

1978; Vol. 3, pp 91-127.

1980, 34, 200-206.

Rinehart and Winston: New York, 1988. (20) Eaton, W. S. Ph.D. Thesis, Unlversity of Wlsconsln-Madison, 1974.

RECEIVED for review September 13,1979. Resubmitted May 7,1981. Accepted June 22,1981. Support for this work was received from the Department of Chemistry, the Graduate School of the University of Wisconsin, and the National Science Foundation under Grant No. CHE-77-05294 and CHE-79-15145. Portions of the work were presented at (1) ACS Joint 10th Central-12th Great Lakes Regional Meeting, May 24-26, 1978, Indianapolis, IN, (2) 5th Annual Meeting of FACSS, Oct 30-Nov 3, 1978, Boston, MA, and (3) 22nd Rocky Mountain Conference, Aug 11-14,1980, Denver, CO.

Overheating Fault Detection in Hydrogen-Cooled Steam Turbine Generators by Chemical Tagging and Gas Chromatography

Woodfin V. Ligon, Jrnq* and Jimmy L. Webb

General Electric Company, Corporate Research and Development, P.O. Box 8, Schenectady, New York 1230 1

An analytical system based on chemlca! tagging compounds has been developed for use In overheatlng fault detection In hydrogen-cooled large !steam turbine generators. The tagging materials are imldes aind an amic acld derived from tetra- chlorophthalic anhydrlde and may be detected in smoke particles using a gas clhromatograph equlpped with an elec- tron capture detector. The unique feature of this work is the successful design of a !set of thermally stable tags which are efflclently Incorporatedl in smoke particles durlng pyrolysis.

Large steam turbine generators are largely responsible for the production of the electric power utilized in the United States. A single machine may produce upwards of 1000 MVA of power and the value of the power produced by such a machine can exceed $500 000 per day. Consequently, it is of unusual importance thLat any faults which might occur be detected and corrected promptly. The problem of detection and location of overheating faults in these machines has been the subject of intensive investigation in a number of labora- tories in recent years. The first approaches to this problem were made by Carson et al. (1) who reported the use of con- densation nuclei detectors and ionization type detectors mounted directly on tho machine to provide an indication of the presence of smoke in the machine. Such detectors are now widespread in the industry and have been shown in numerous cases to reliably provide an indication of overheating.

A number of investigators noted that the pyrolysate products being detected1 by this instrument should be amen- able to chemical analysis in order to determine their origin and thereby elucidate the location of the fault condition given that the layout of the various insulation systems is known from the design of the machine. It has been reasoned that other characteristic pyrolysate products should be present in un- condensed form in the gas phase. Work has been reported in which both sources of potentially characteristic pyrolysates have been investigated (2, 3).

These approaches to fault location have also been inves- tigated in this laboratory. It has been found that investigation of the gas-phase pyrolysate via trapping, for example, on porous polymer adsorbents with subsequent gas chromatog-

raphy-mass spectrometry (GC-MS) analysis suffers from se- vere interferences due to large amounts of volatiles normally present in the hydrogen coolant atmosphere. For this reason we do not consider this approach useful. Trapping of the particulate pyrolysate products on glass fiber filters provides a sample largely free of this indigenous hydrocarbon back- ground. Analysis of these samples via direct thermal de- sorption (DTD) into a GC-MS (DTD-GCMS) provides a sample which is characteristic of the material undergoing decomposition. Unfortunately this approach also has a number of shortcomings. First the exact nature of the py- rolysate products produced from a given polynieric insulation material is a very strong function of temperature. At relatively low pyrolysis temperatures one may observe a given set of products while a t higher temperatures these products may change drastically in ratio or may disappear completely and be replaced with other products.

Secondly, since an essentially continuous range of tem- peratures from 100 OC to the melting point of the metallic substrate must be considered possible, it is very difficult to predict the product mixture reliably. Accordingly, it is nec- essary that a t least one pyrolysis product be found that is completely unique to each material which must be discrim- inated. Simple differences in ratios of products are largely irreproducible and not useful. In our experience certain materials can in fact be discriminated on the basis of unique products. For example, certain 2,2-bis(4-hydroxyphenyl)- propane (BPA) epoxies are characterized by the presence of BPA as a pyrolysis product. Unfortunately a given generator may contain as many as a dozen slightly different BPA epoxies in critical locations all of which can give BPA on heating. Therefore this approach while providing some information about the nature of the substrate undergoing decomposition is not useful for discriminating the large number of relatively similar materials in a generator and thereby locating the site of overheating.

Since it has not been found possible to discover pyrolysate products unique to all of the areas of interest, in a generator, it becomes necessary to impose such a distinction on the materials artifically via some tagging scheme. Tagging pro- grams cannot be entered into lightly. Generator insulation systems have historically been the subject of unusually ex-

0003-2700/81/0353-1655$01.25/0 0 1981 American' Chemical Society

1656 ANALYTICAL CHEMISTRY, VOL. 53, NO. 11, SEPTEMBER 1981

tensive and painstaking research programs in order to discover materials which have the necessary combination of chemical, electrical, and mechanical properties. Rigid specifications are necessarily placed on each component and changes are made rarely and only after extensive testing. Clearly then the ad- dition of a foreign component into these systems must be extensively justified and can in no way compromise the performance of the insulation system or introduce any po- tential hazard to the remainder of the machine.

This problem has been approached by other workers. Smith et al. (4) have described coating compositions which both give unique products on pyrolysis and further tend to particulate at temperatures somewhat lower than the resin which they are intended to tag thereby potentially giving early warning of overheating conditions. These compositions would seem to suffer from two inherent weaknesses. First, the preferred compositions described suggest percentages as high as 50 for the active agent. It seems unlikely that such a composition could be utilized without significantly altering the properties of the insulation system. Second, since the normal projected life of a generator is at least 40 years, there would appear to be potential problems related to useful lifetime for the nec- essarily relatively unstable materials being used to provide the early warning function.

The present authors have approached this problem via chemical tagging of existing insulation systems a t relatively low levels followed by pyrolysate analysis using gas chroma- tography with electron capture detection.

EXPERIMENTAL SECTION The tagging compounds were synthesized by reacting primary

and secondary amines with tetrachlorophthalic anhydride using conventional techniques. The compounds used have general formulas I and I1 where R1 = R2 = n-hexyl and R3 = n-hexyl, n-octyl, cyclooctyl, n-dodecyl, and adamanty1. These compounds

I I1 are compounded uniformly into a two-part epoxy paint at the 0.1-0.5% level. The paints are also color coded in order to help ensure that each tagged paint will be applied in the correct generator location.

These tagged paints have been evaluated in the following manner. Nichrome strips approximately 1 in. wide by 12 in. long were painted individually with paints tagged with each of the tagging compounds both singly and in combination. These strips were mounted inside an actual operating generator such that they could be successively made part of a circuit energized by a 400-A arc welder power supply. When power was applied to a given strip, the paint was pyrolyzed producing a tagged smoke. This smoke was collected on a glass fiber disk (13 mm diameter, Millipore Filter Corp.) starting as soon as a pyrolysate signal appeared on the generator's ion chamber detector. Smoke sampling was continued as long as the pyrolysate signal remained. Smoke sampling involves venting a small fraction of the generator's hydrogen gas coolant through the glass filter. In general about 100 ft3 (measured at atmospheric pressure) of the generator's total of about 5000 ft3 (45 or 60 psig) was sampled for each test specimen.

The glass fiber disk containing pyrolysates were subsequently treated with 100-500 pL of Nanograde (Mallinkrodt Chemical) methanol. Samples of 1 pL were injected into a gas chromatograph for analysis.

The gas chromatographic conditions were as shown in Table I.

The gas chromatograph was calibrated by using a standard mixture of known tagging compounds. Aliquots of the unknown pyrolysate solutions obtained by methanol treatment of the filter disks were then injected. The tagging compounds present were identified by retention time with reference to the standard. Under

Table I. Gas Chromatographic Conditions instrument column column packing

oven temp

injector temp detector temp attenuation slope sensitivity carrier gas flow detector

Hewlett-Packard 5830A l/4-in. 0.d. X 2-mm i.d., 6-ft Pyrex 3% OV-17 on Gas Chrom Q (Applied

programmed from 240 to 280 "C at 5

300 "C 305 "C

0.1 methane-argon 25 mL/min 63Ni electron capture

Science Laboratory, Inc.)

"C min-I

27

these GC conditions, the retention times are reproducible within 0.04 min or better.

DISCUSSION AND RESULTS There are a number of criteria which any substance being

considered as a tagging compound must meet. These may be outlined as follows:

(1) The tag must be easily detected and identified in the presence of a very large number of pyrolysis products.

(2) The sensitivity of the combined analytical system (tag and detector) should make possible the addition of very low levels of tag to minimize any chance of the tag altering the characteristics of the insulation system.

(3) The tag should be free of any properties which might adversely affect operation or lifetime of the generator such as, for example, the initiation of stress corrosion cracking. (4) The tag should be essentially nonvolatile at normal

operating temperatures in order to remain in place over the lifetime of the generator.

(5 ) Finally, it was concluded that tags should be sought which would be successfully incorporated in smoke particles. This was decided for several reasons: First, experience with trapping of volatile products had found this approach to be frought with interferences and subject to adsorption and desorption efficiencies. Second, the smoke particle is unam- biguously derived from the overheated part whereas volatile pyrolysis products are subject to dilution by indigenous volatiles. Third, i t is the smoke particle which is detected by the ion chamber detector. Fourth, it appeared more difficult to design tags which would both remain in the gas phase and have long lifetimes per point 4 above. To be trapped by a smoke particle, a tag will need very good thermal stability so as to survive the pyrolysis and low volatility so as to be trapped in the smoke particle.

The analytical system finally selected consists of tagging compounds chosen from a series of N,N-dialkyltetrachloro- phthalamic acids and N-alkyltetrachlorophthalimides in combination with a gas chromatograph equipped with an electron capture detector. The compounds being used are shown in the Experimental Section. The analytical system satisfies the conditions set forth above in the following manner: First the compounds can be easily resolved and identified by a gas chromatograph using an electron capture detector. Second, the detector is sufficiently sensitive that as little as 100 pg of tag can be easily and unambiguously detected. This means that the tag can be incorporated a t low levels in the insulation. Third, these tagging materials are quite stable thermally, tending to distill (even at temperatures over 400 T) rather than thermally decompose. For this reason it has been found that these materials can be safely incorporated in the generator with no fear that will librate free ions which might initiate stress corrosion cracking. At the levels used, no other negative effects of the tags have been identified. Fourth, for the alkyl groups used, the tagging compounds have very low volatility. As will be noted later, the minimum

ANALYTICAL CHEMISTRY; VOL. 53, NO. 11, SEPTEMBER 1981 1857

Kovats index (5) on OV-17 for a tag now in use is 2400. It is believed, therefore, that when included in a polymer matrix, these tags will have acceptable lifetimes in the generator. Fifth, the materials selected are effectively incorporated in smoke particles (vide infra).

The tagging compounds chosen for final application were screened from a list of over 50 members of the two classes of compounds which were synthesized. The first criteria for selection was that the materials be easily resolved on the gas chromatograph. After this condition was met the resolved sets of compounds were submitted to testing in an actual operating generator as described in the Experimental Section.

A number of compounds were eliminated at this stage be- cause they could not bo detected in smoke particles generated from overheated insulation. It was noted that, in general, the more volatile derivatives (smaller alkyl group) failed in this experiment. Accordingly, a second competitive experiment was conducted in which all of the tags under consideration both good and bad varying over a large moleculllr weight range were compounded at equal levels in a single insulation system. When this sample was 13ubmitted to pyrolysis in the generator, it was again observed that the more volatile components were apparently not incorporated in smoke particles and could not be detected. This unexpected discover? can be generalized as follows: For the combination of dialkyltetrachloro- phthalamic acids or alkyltetrachlorophthalimides, and epoxy paint (vide infra), the tag will not be efficiently incorporated in the smoke paiticles if the Kovats index on OV-17 is <2400. The authors propose the following explanation of this phe- nomena: Smoke particles which are produced on pyrolysis in generator can be envisioned as forming in an area just above the heated surface. The temperature a t this poidt is neces- sarily lower than that of the surface and the heavier materials being volatilized from the surface can proceed to condense to small droplets or smoke. This must occur very near the surface so that the volatiles are not sufficiently diluted by the hydrogen atmosphere to prevent agglomeration. It can be reasonably assumed that only pyrolysis products with boiling points as high or higher than the temperature in this con- densing zone will be incorporated into smoke particles. More volatile materials could of course be adsorbed later a t lower temperatures further from the surface, but the efficiency of adsorption would suffer due to dilution effects. The sharp cutoff in volatility for an effective tagging compound suggests that the boiling point of the lowest member of the set (equivalent to about a C2* hydrocarbon) represents roughly the temperature of the condensing or smoke producing zone in these experiments.

Experiments in operating generators also included studies in which both the tagging compounds and the substrate in- sulation system were variables. The nature of many of the insulation materials tested is proprietary, however, certain conclusions of a general nature drawn from this work can be mentioned. First the tagging compounds were not found to be equally effective when incorporated in differing substrates. In particular, the tags were found to be completely ineffective in insulation systems which underwent processing at tem- peratures over about 250 "C. One of the most effective tag carriers proved to be a two-part epoxy paint which is normally used as a final protective interior overcoat just prior to shipment. This paint has been made part of the overall tagging system.

The sensitivity of the tagging system when utilized in an operating generator may be evaluated semiquantitatively in several ways. First machine tests described in the Experi- mental Section showed that a quantity of smoke which re- sulted in a detectable change in the signal from the ionization type overheating detector would provide a large enough

TAGGING COMPOUND REFERENCE MIXTURE

w v) z

ul W LT

LT 0 c u W

c- W

B

a

I

T I M E

Flgure 1. Electron capture-gas chromatogram of 6 chemical tagging compounds: (3.71) N-n-hexyltetrachlorophthalimide, (4.16) N,Ndi- n-hexyltetrachlorobenzamide, (5.37) N-n-octyltetrachlorophthallmide, (7.95) N-cyclooctyltetrachlorophthalimide, (9.95) N-n-dodecyltetra- chlorophthalimide, ( 12.44) N-( I-adamantyl)tetrachlorophthalimlde.

I SMOKE SAMPLE

I T I M E

Figure 2. Electron capture-gas chromatogram of a tagged pyrolysate sample.

particulate sample for tag identification when 100 ft3 of coolant was sampled. This means that an overheating incident which results in an alarm level of smoke in the machine will provide a much more than adequate sample for gas chromatography.

The smallest area of overheating which can be located has not been evaluated rigorously, however, we have recently experienced a failure in an operating machine in which dro- plets of molten copper fell onto an area which was coated with N-(cyclohepty1)tetrachlorophthalimide-tagged paint. The affected area was less than 1 in.2 in area but nevertheless analysis of a smoke sample allowed unambiguous identification of the tag based on a small but definitive GC detector re- sponse. It should be emphasized that in all cases GC analyses are carried out on crude smoke samples with no attempt a t prepurification and that in fact only 1/100 to 1/500 of the total sample collected is used for the determination. It is reasonable to assume, therefore, that cleanup procedures analogous to those routinely used for pesticide residue analysis would greatly enhance the sensitivity should this ever be needed.

The overall tagging system functions as follows: Samples of paint are uniquely tagged and an identifying pigment is added. The tagged paints so obtained are applied selectively to the various parts of the generator which may encounter overheating. Since the paints are normally a final coating,

1658 Anal. Chem. 1981, 53, 1658-1662

they may be applied to both new units and units already in service. When a part of the generator overheats, this neces- sarily results in pyrolytic decomposition of the tagged coating in that area, with production of tagged smoke.

This smoke is detected by an ion chamber detector which sounds an alarm and automatically starts sampling of the hydrogen coolant through a glass fiber particulate filter via activation of a solenoid valve. This filter is subsequently separated from its holder and analyzed as described in the Experimental Section. Once the tag is identified, this in- formation can be compared with the tagging key for that generator and the location of the overheating thereby de- termined.

Figure 1 is the electron capture-gas chromatogram of a standard containing all of the tags presently proven useful. Figure 2 is the electron capture gas chromatogram of a methanol extract of a pyrolysate sample from an actual op-

erating generator. The tag was identified as N-(n-octy1)- tetrachlorophthalimide based on its retention time.

ACKNOWLEDGMENT The authors wish to acknowledge considerable assistance

provided during this work by C. C. Carson, F. S. Echeverria, R. S. Gill, M. Cipullo, and L. Burks.

LITERATURE CITED (1) Carson, C. C.; Barton, S. C.; Grobel, L. P. IEEE, PES Winter Meetlng,

Jan 1971, Paper 71CP 154-PWR. (2) Kelly, J. K.; Aulk. V. J.; Herter, V. J.; Hutchinson, K. A.; Rugensteln, W.

A. IEEE, PES Winter Meeting, Jan 1974, Paper T74-218-4. (3) Sexton, R. M.; Dillman, T. L. Forty-Second Annual International Con-

ference of Doble Clients, Boston, MA, April 1975. (4) Smlth, J. D. 8.; Phillips, D. C.; Kaczmarek, T. D. Anal. Chem. 1976,

48, 89. (5) Wehrli, A.; Kovats, E. Helv. Chim. Acta 1959, 42, 2709.

RECEIVED for review March 2,1981. Accepted June 22,1981.

Reversed-Phase Ion-Pairing Liquid Chromatographic Separation and Fluorometric Detection of Guanidino Compounds

Mark D. Baker,' Hussain Y. Mohammed,2 and Hans Veening"

Department of Chemistry, Bucknell University, Le wisburg, Pennsylvania 17837

Several guanldlno compounds are separated by Ion-pairing reversed-phase liquid chromatography with an isocratic mo-

Table 1. Guanidine Structures

bile phase consisting of aqueous-acetate buffer and methanol. The mobile phase also contains hexanesulfonate (HSA-) as the counterion. On-line postcoiumn derivatiration of the guanldlnes with phenanthrenequinone (PQ) in an alkaline stream precedes fluorescence detection. Experiments are carried out to determine the effects of several variables such as pH, concentration of the aqueous buffer, concentration of the ion-pairing agent (HSA-), and the percentage of methanol in the movlng phase. Working curves are llnear in the range of 0.05-2.5 nmol. This method is applied to normal and uremic physiological fluids.

Guanidine (G) and its monosubstituted derivatives, me-

R -CH,COOH

-CH ,CH ,COOH

-CH,CH,CH,COOH

-CHCOOH I CH,COOH

-CH ,CH, SO;

-H

-CH,

H -(cH,),-L-cooH thylguanidine (MG), guanidinosuccinic acid (GSA), and others,

are highly suspect as potential uremic toxins in renal patients (1-8). The structures of these and other guanidino compounds are presented in Table I. The concentration of these com- pounds in uremic and normal physiological fluids has been used for some time as an indicator of renal dysfunction (%12). The availability of a rapid and sensitive method for the de- termination of guanidine levels in biofluids would thus rep- resent a potentially useful clinical assay procedure.

Previous methods for the separation and detection of guanidino compounds have included classical column chro- matography using ion-exchange resins coupled with colori- metric detection (1 ,3 , 6, 9, 10-14) or fluorometry (2, 11) as well as automated ion-exchange with colorimetric detection

Present address: E. I. du Pont de Nemours & Co., Inc., Aiken,

aPresent address: E. I. du Pont de Nemours & Co., Inc., Belle, SC 29808.

WV 25015.

NH,

name (abbrev) guanidinoacetic acid (GAA) 3-guanidinopropionic acid (GPA) 4-guanidinobutyric acid (GBA)

guanidinosuccinic acid (GSA)

taurocyamine (TC) guanidine (G) methylguanidine (MG)

arginine ( ARG)

FH'

creatinine (CRN) H\ /N\c=N", "'i I dJ-"

15-1 7). Paper chromatographic methods using colorimetric detection (2, 6, 18-20) or fluorometry (21J have also been reported. In addition paper electrophoresis with colorimetric detection has been used to identify guanidine compounds (1, 3,9). Numerous gas chromatographic separations for guan- idines have also been developed (22-28). Methylguanidine has been selectively separated from physiological fluids as its picrate ion-pair using liquid-liquid column chromatography coupled with UV detection (29). Recently, a high-performance liquid chromatographic (HPLC) ion-exchange method using

0003-2700/81/0353-1658$01.25/0 0 1981 Amerlcan Chemical Soclety