International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance...

64
International Journal for Ion Mobility Spectrometry 2 (1999)1 Official publication of the International Society for Ion Mobility Spectrometry I N T E R N A T I O N A L S O C I E T Y f o r I O N M O B I L I T Y S P E C T R O M E T R Y

Transcript of International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance...

Page 1: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

International Journal

for Ion Mobility Spectrometry

2(1999)1

Official publication of the

International Society for Ion Mobility Spectrometry

INTERNATIO

NAL

SOCIE

TY for ION MOBILITYSPECTROMETRY

Page 2: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers
Page 3: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

Table of Contents

J.I. Baumbach, P. Pilzecker, E. Trindade

35Monitoring of Circuit Breakers using Ion Mobility Spectrometry to detectSF6-Decomposition

Ion mobility spectrometry (IMS) has been known as an analytical technique since the late 1960s and early 1970s. Todate, it has been successfully utilized for the detection of environmental pollutants, warfare agents, explosives, herbi-cides, pesticides, petroleum products as well as for the detection of prescription and illicit drugs. In this paper theauthors describe the first IMS field tests in Germany and Austria. We also report the use of the IMS methodology forthe rapid analysis of hallucinogenic fungi material. A new application of the IMS technology for the analysis of postmor-tem sweat samples for drugs is also presented.

Th. Keller, A. Schneider, E. Tutsch-Bauer, J. Jaspers, R. Aderjan, G. Skopp

22Ion Mobility Spectrometry for the Detection of Drugs in Cases of Forensicand Criminalistic Relevance

In the case of investigations of mixtures of analytes using ion mobility spectrometers (IMS) peak overlapping and additi-onal peaks, occurring because of dimers containing different molecules of the analytes, often reduce the resolution ofthe spectra. Therefore, the influence of inter-molecular charge transfer reactions on the signal should be reduced bytime delayed sample introduction into the ionization regions of the IMS realized by coupling Multi-Capillary columns(MCC) to the IMS. The aim of this combination is to achieve further enhancement of the resolution of the instrumentand to significantly increase the scope of application of ion mobility spectrometry. The response of an UV - ion mobilityspectrometer (IMS) to trans-1,2-dichloroethene, trichloroethene and tetrachloroethene and the advantages of thecombination of multMCC with ion mobility spectrometers will be discussed. The advantages of MCC-IMS compared toMCC coupled to a photoionization detectors will be demonstrated by achieving efficient separation of mixtures atambient temperature within about 1 minute. The results are discussed with respect to the linear range and theminimum detectable limit of the 10.6 eV - UV - ion mobility spectrometer particularly in regard on field applications.

S. Sielemann, J.I. Baumbach, P. Pilzecker, G. Walendzik

15Detection of trans-1,2-dichloroethene, trichloroethene and tetrachloroetheneusing Multi-Capillary Columns Coupled to Ion Mobility Spectrometers with

UV-Ionisation Sources

Corona discharge ion mobility spectrometry permits the determination of n-alkanes and branched chain alkanes withinthe low ppm range. For n-alkanes, one or two product ion peaks are observed. Two product ion peaks are detectablein lower concentration range only (<30-45 ppm in dependence on physical and chemical properties of n-alkanes). Withincreasing concentrations, the ion mobility spectra show one product ion peak with a sharp peak profile. The reducedmobilities of these peaks decrease regularly with increasing molecular weight. Series of product ion peaks are obtainedfor branched chain alkanes. Characteristic fragment ions are additionally formed in comparison with n-alkanes.

H. Borsdorf, H. Schelhorn, J. Flachowsky, H.-R. Döring, J. Stach

9Determination of n-alkanes and branched chain alkanesby Corona discharge ion mobility spectrometry

Many of the parameters that affect negative electrospray atmospheric pressure ion mobility spectrometry were investi-gated. They included solvent composition, temperature, spray voltage, solvent flow rate, drift gas flow rate and coolinggas flow rate. Using a solvent of methanol and water it was found that the signal intensity of our test compound,adenosine 5-monophosphate (AMP) decreased with increasing water concentration, showing no signal at water concen-trations greater than 40%. The total ion current (TIC) showed a similar response at low water concentrations,however at higher water concentrations the TIC began to increase again probably due to formation of a coronadischarge from the needle tip. Increasing the temperature increased the signal intensity of AMP and the TIC. Thetemperature also influenced the resolving power measured for AMP, reaching a maximum at 150 °C. Higher sprayvoltages generally showed higher TIC, however AMP signal intensities reached a maximum and then began to decrease.At voltages higher than this maximum, a corona discharge ionization was probably present. Signals were observed forAMP at solvent flow rates from 2mL/min to 100mL/min, with a maximum signal occurring at 4mL/min. No significantresponse changes were noted for changes in drift gas flow rate or cooling gas flow rate. In addition to methanol threeother organic modifiers were studied including ethanol, 2-propanol, and acetonitrile. In all cases improved electrosprayperformance was shown for solvents with high concentrations of organic modifier and low concentrations of water.

G. Reid Asbury and Herbert H. Hill, Jr.1Negative Ion Electrospray Ionization Ion Mobility Spectrometry

Papers

Copyright © 1998 by International Society for Ion Mobility Spectrometry

Page 4: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

61Newsletter of the International Society for Ion Mobility Spectrometry

The training of neural networks is assessed using several terms and these lack precision with large libraries of ionmobility spectra since the terms are averaged for the total number of training spectra. Consequently, the measures failto show when a network has reached an optimum in the extent of training and the decision to halt network trainingbecomes arbitrary and prone to error. A numeric and graphic assessment of training by a neural network is possiblewhen network performance is treated as a Fermi-Dirac distribution for an extensive library of mobility spectra. Thisnew method provides refined measures of network performance or status. In addition to providing clear criteria toprevent over-training (with large savings in both time and committed computational resources), the graphic method isalso helpful in identifying flawed mobility spectra in training sets. Modeling of distributions for network performancewas also used to identify the regions of high information content in mobility spectra. The region near the reactant ionswas found to contain information necessary for successful training with mobility spectra. This unexpected result maydisclose details yet unknown regarding gas phase ion chemistry in ion mobility spectrometry.

Erkin Nazarov, G.A. Eiceman, Suzanne E. Bell

45Quantitative Assessment for the Training of Neural Networks with LargeLibraries of Ion Mobility Spectra

Positive ion mobility spectra of chlorobenzene, bromobenzene and iodobenzene were measured using different ioniza-tion processes. Corona discharge ionization and photoionization permit a more sensitive detection of thesecompounds in comparison with 63Ni ionization. For chlorobenzene and bromobenzene, a similar drift behavior offormed product ions can be observed. Corona discharge ionization and photoionization provide one product ion peak.However, the reduced mobility values (K0 values) vary depending on the used ionization method. For 63Ni ionization,two product ion peaks are detected for these compounds. The reduced mobility values of these two peaks correspondto K0 values of the peaks observed by corona discharge ionization and photoionization. For iodobenzene, two peakscan be obtained for all ionization processes. However, the most intensive peak is detected at the same reduced mobil-ity values for all ionization processes.

H. Borsdorf, H. Schelhorn, M. Rudolph, J. Flachowsky, J. Stach

40Ion mobility measurements of mono-halogenated benzenes using differentionization processes

In high-voltage systems insulated with SF6 electrical discharges as partial discharges, sparks or arcs cause SF6 decompo-sition leading to the formation of some toxic and corrosive by-products. There is an urgent need for more informationon the origins and quantities of contaminants expected to arise from the use of SF6 filled electrical power equipment. Alow resolution ion mobility spectrometer used as analysis instrument delivers a shift of the position of the main peakobtained. This shift is correlated to the concentration of decomposition products formed in SF6. This paper presentsthe results of investigations on the fill gas in circuit breakers in gas insulated substations during operation and of thereclaimed gas after a recycling procedure. Results from the investigation of 36 different circuit breakers in an operatingsubstation are presented, which can lead to new methods to check the fill gas quality.

Copyright © 1998 by International Society for Ion Mobility Spectrometry

Page 5: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

AbstractMany of the parameters that affect negativeelectrospray atmospheric pressure ion mobilityspectrometry were investigated. They includedsolvent composition, temperature, sprayvoltage, solvent flow rate, drift gas flow rateand cooling gas flow rate. Using a solvent ofmethanol and water it was found that the signalintensity of our test compound, adenosine5-monophosphate (AMP) decreased withincreasing water concentration, showing nosignal at water concentrations greater than40%. The total ion current (TIC) showed asimilar response at low water concentrations,however at higher water concentrations theTIC began to increase again probably due toformation of a corona discharge from theneedle tip. Increasing the temperatureincreased the signal intensity of AMP and theTIC. The temperature also influenced theresolving power measured for AMP, reaching amaximum at 150 °C. Higher spray voltagesgenerally showed higher TIC, however AMPsignal intensities reached a maximum and thenbegan to decrease. At voltages higher thanthis maximum, a corona discharge ionizationwas probably present. Signals were observedfor AMP at solvent flow rates from 2µL/min to100µL/min, with a maximum signal occurring at4µL/min. No significant response changeswere noted for changes in drift gas flow rate orcooling gas flow rate. In addition to methanolthree other organic modifiers were studiedincluding ethanol, 2-propanol, and acetonitrile.In all cases improved electrospray performancewas shown for solvents with highconcentrations of organic modifier and lowconcentrations of water.

IntroductionElectrospray ionization was first introduced asan ionization source for ion mobilityspectrometry in 1972 by Dole and co-workers[1]. Though these initial experiments did leadto spectra of lysozyme, the peaks were very

broad, most likely due to inefficientevaporation. Later Dole concluded thatbecause of the difficulty in removing thesolvent, electrospray would only be useful asan ionization source for mass spectrometryunder vacuum conditions [2]. Similar resultswere observed by Smith and co-workers in1991 [3]. Although the majority of research intoelectrospray ionization shifted to massspectrometry in the late 1980’s and 1990’sfollowing the pioneering work of Fenn andco-workers [4], there were attempts to use heatand unidirectional flow to desolvate ions foratmospheric pressure ESI-IMS [5]. Theseworks were successful at analyzing lowmolecular weight and semi volatile compounds,but proved ineffective at ionizing highmolecular weight compounds. Hill andco-workers found that by cooling theelectrospray unit one could obtain partiallydesolvated ions of high molecular weightspecies by ESI-IMS at atmospheric pressure[6]. It was thought that heat was needed todesolvate ions in the spectrometer, but localcooling of the ESI source was necessary tokeep the solvent from boiling prior to exitingthe spray needle. Other groups used partialvacuum to desolvate ions for IMS [7]. Thishowever, limited the pressure in which themobility measurements could be madeultimately limiting the resolving power possible.Other designs were demonstrated byGuevremont et al [8], in which ions weresprayed into a capillary at atmosphericpressure. Finally completely desolvatedelectrosprayed ions at atmospheric pressurewere recently observed by Hill and co-workers[9]. This new design allowed theelectrosprayed ions to spend more time in theheated desolvation region before they wereintroduced into the drift portion of thespectrometer. With this design, baselineresolved charge states were observed for bothcytochrome C and ubiquitin.

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Negative Ion Electrospray Ionization Ion Mobility Spectrometry

G. Reid Asbury and Herbert H. Hill, Jr.

Department of Chemistry, Washington State University, Pullman, WA 99164-4630

Received for review December 2, 1999, Accepted December 14, 1999

Page 6: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

Despite the extensive work done in the positivemode there has been virtually no systematicstudy of negative mode electrospray foratmospheric pressure IMS. Only a fewcompounds have ever been published usingnegative mode electrospray at atmosphericpressure [10], all of which are veryelectronegative acids. Negative modeelectrospray mass spectrometry has beenstudied since the pioneering work of Fenn andothers [11], however, only a few extensivestudies of negative mode electrospray formass spectrometry exist [12]. One of thebiggest obstacles to negative electrospray isthe tendency of the source to form a coronadischarge probably created by electrons beingemitted from any sharp edges on theelectrospray needle. In mass spectrometry anumber of solutions to stop formation of acorona discharge have been used includingadding electron scavenging gases around theneedle [11,13] and using non-aqueoussolvents [14]. The same fundamental studiesusing atmospheric pressure ion mobility andheat for desolvation have been non-existent.

In this paper we report the first parametricstudy of negative ion electrospray atmosphericpressure ion mobility spectrometry. A numberof parameters were studied including solventcomposition, temperature, spray voltage,solvent flow rate, cooling gas flow rate, anddrift gas flow rate.

Experimental

InstrumentationAll data was collected using an electrosprayion mobility spectrometer / mass spectrometerconstructed at Washington State University.The electrospray source was a speciallydesigned water-cooled device. Theelectrospray needle was housed in awater-cooled jacket and was additionallycooled with water cooled nitrogen gas blowingalong the axis of the needle. This cooling keptthe needle below 50 °C despite being in closeproximity to a 250 °C spectrometer. Theelectrospray source has been describedpreviously [6a].

The ion mobility spectrometer consisted of tworegions, a desolvation region and a drift region.The desolvation region consists of the first

7.2cm of the spectrometer with the drift regioncomprising the last 13cm of the spectrometer.Both regions are constructed of electricallyconducting stainless steel and electricallyinsulating alumina rings stacked in analternating and interlocking design. The firstring of the desolvation region is a 16-meshstainless steel focus screen. An ion gateseparates the desolvation region from the driftregion. The ion gate is a Bradbury-Nielsentype ion gate constructed of parallel wires(California Fine Wire Co., Grover Beach, CA)76µm in diameter with 0.64mm spacings. Thegate can be closed by applying a potential ofabout 25V to adjacent wires creating anelectric field of about 700V/cm orthogonal tothe drift field of the spectrometer. Thiseffectively allows no ions to pass. When thegate is open both wires are referenced to theappropriate voltage to the gates position in thespectrometer. All other details of theinstrumental setup have been describedpreviously [9].

General operating conditionsThe entire spectrometer was heated tobetween 100 and 250 °C and the drift gas waspre-heated to the same temperature as thespectrometer. Two high voltage powersupplies were used, one was used to apply apotential to the drift tube, the other to apply apotential to the electrospray unit. The first ringof the drift tube was held at –6000V while theelectrospray needle was adjusted from –7500to –11,000 V providing a difference of between–1500 and –5000V relative to thespectrometer. The needle was positioned inthe center of the spectrometer. The position ofthe needle was further adjusted to provideoptimal signal intensity (for AMP) under eachset of conditions. The solvent flow ratethrough the needle was between 2-100µL/minand was maintained by a dual piston syringepump (Brownlee Labs, Santa Clara, CA.). Allspectra were taken using a data acquisitionsystem designed some time ago at WSU [6].The spectra shown were the average of 256individual spectra. The gate was held opencontinuously when the TIC was monitored. The occurrence of a corona was determined bymonitoring the solvent ions produced.

G.R. Asbury and H.H. Hill: „Negative ion electrospray ...”, IJIMS 2(1999)1, 1-8, p. 2

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Page 7: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

Reagents The solvents were purchased from J.T. Baker(Phillipsburgh, NJ) and were listed as reagentgrade or better. The adenosine5-monophosphate was purchased from SigmaChemical Company (St. Louis, MO) and wasused without purification. A stock solution ofAMP at 1 mg/mL was made in pure water.This stock solution was then diluted to0.01mg/mL in the appropriate solvent.Therefore all solutions did contain a smallpercentage of water.

Results and discussionMethanol and water solventsThe effects of using varyingconcentrations of water in methanolon negative electrospray IMS wereinvestigated. Figure 1 (A) shows theTIC for the solvent at differentcompositions of water and methanol.At 100% methanol the total ioncurrent was about 1.3nA, this isabout half of the typical ion current inthe positive mode (3nA). At 75:25(v/v) methanol and water there was adecrease in total ion current to about1.2nA. At equal volumes of waterand methanol the total ion currentincreased to over 1.5nA. Figure 1 (B)shows the signal intensity for AMP inthe same solvents as above. In thiscase there was a 60% decrease insignal in going from 100% methanolto 75% methanol. In addition, nosignal was seen for AMP in the 50%methanol solution. These resultsprobably indicate that at 50%methanol a corona discharge hasformed and true electrosprayionization was no longer occurring.This explains both the increased TICand the lack of signal for AMP. AMPis a very polar non-volatile compoundthat is likely insensitive under coronadischarge conditions. Furtherevidence of corona dischargeionization in the 50% methanolsolution can be seen in Figure 2 (Aand B). These two figures show themobility spectra of the 100%methanol solution and the 50%methanol: 50% water (v/v) solution.

In the 100% methanol solution there are atleast five ions seen, chloride, nitrite, formate,nitrate, and acetate. All of the ions wereidentified by mass spectrometry. The 50%methanol solution shows the same five ions,but the intensities of the ions are muchdifferent. Both the nitrite and the nitrate ionsare much more intense while the other ions arereduced in intensity. Nitrite and nitrate arebelieved to be formed in the gas phase at thecorona discharge. With our current system,solutions of more than 10% water in methanolshow much decreased signal intensity for AMPand solutions of greater than 40% water show

G.R. Asbury and H.H. Hill: „Negative ion electrospray ...”, IJIMS 2(1999)1, 1-8, p. 3

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 1:The top graph (A) shows the TIC for solutions withdifferent concentrations of water in methanol usingnegative mode electrospray ionization ion mobilityspectrometry. The bottom graph (B) shows the signalintensity observed for AMP using these same solutions.All of the data was taken at solvent flow rate of 5µL/min, and a temperature of 250 °C. The spray voltage usedwas that which provided the maximum signal.

0

100

200

300

400

500600

700

Sig

nal I

nten

sity

(A

MP

)

455565758595105

Methanol %

Figure 1

11501200125013001350

14501500

15501600

1400

TIC

(pA

)

455565758595105

Methanol %

A

B

Page 8: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

no signal for AMP. Since it is now routine inmass spectrometry to spray solutions of 50%water or greater in the negative mode thedifference must be in either our in house builtelectrospray unit or in the rigorous conditionsrequired to desolvate the ions without the useof vacuum. Temperature effectsThe effect of temperature (100 °C to 250 °C)on TIC, signal intensity of AMP, and on theresolving power measure for AMP using 100%methanol was determined. Figures 3 (A and B)show plots of temperature versus TIC andsignal intensity of AMP. The plots show that asthe temperature is increased the TIC and thesignal intensity of AMP increase. The shape ofthe curves is considerably different however.

The TIC increases rapidly until about 180 °Cand then levels off whereas the signalintensity of AMP rises slowly and then beginsto rise rapidly above 180 °C. The overall trendof increased signal at increasing temperatureis probably a result of decreased ionclustering. Another reason for increasedsignal intensity is probably due to increasedion velocity at higher temperatures. Thisincreased velocity is due to both thedecreased clustering and the normal increasein mobility at higher temperatures. As the ionvelocity increases the total time spent in thespectrometer decreases. This decreased timemay help eliminate any losses do to gateeffects and losses due to ion moleculereactions in the spectrometer. Similar resultshave commonly been seen for traditional IMSusing 63Ni ionization.

The resolving power for AMP at differenttemperatures was also measured and isshown in figure 4. The sharp increase inresolving power in going from 125 °C to 150

G.R. Asbury and H.H. Hill: „Negative ion electrospray ...”, IJIMS 2(1999)1, 1-8, p. 4

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 2:Spectra (A) shows the ion mobility spectra of100% methanol (solvent only) and spectra Bshows the mobility spectra of a 50% methanol:50% water solution (v/v). The spectra weretaken at 250 °C at a flow rate of 5µL/min. Thespray voltage for the 100% methanol solutionwas –2700V and the spray voltage for theequal mixture was –4500V

100% Methanol

chlo

ride

nitr

ite

form

ate

nitr

ate

acet

ate

50% Methanol

chlo

ride

nitr

ite

form

ate

nitr

ate

acet

ate

0

200

400

600

800

1000

1200

Inte

nsity

6 7 8 9 10 11 12Drift Time (ms)

0

200

400

600

Inte

nsity

6 7 8 9 10 11 12Drift Time (ms)

Figure 2

B

A

Figure 3:Graph A shows the effect of temperature onTIC while graph B shows the effect oftemperature on AMP signal intensity. All datawas collected using 100% methanol at5µL/min solvent flow rate and a spray voltageof –2700V.

100

200

300

400

500

600

700

Sig

nal I

nten

sity

(A

MP

)

Temperature

90 140 190 240

1050

1100

1150

1200

1250

1300

1350

TIC

(pA

)

1400

90 140 190 240

Temperature

Figure 3

A

B

Page 9: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

°C probably indicates this is thetemperature required to completelydesolvate AMP ions. Furtherincreases in temperature only increasethe broadening due to normal diffusionand therefore do not help increase theresolving power. In previous work inthe positive mode it was found that atemperature 250 °C was required tocompletely desolvate positive ions [8].The difference is likely due to thesolvent differences. In this study100% methanol was used while in theother study a solution of 47.5% water:47.5% methanol: 5% acetic acid wasused. In addition this study onlyexamined one compound, AMP.Other compounds may require highertemperatures to complete desolvation.

Spray voltageThe effects of spray voltage on TICand signal intensity of AMP wasinvestigated using 100% methanolsolvent and a temperature of 250 °C.A minimum voltage of –1700V wasrequired to produce any ions. Figure5 (A and B) shows the effects of sprayvoltage on TIC and AMP signal. Asthe spray voltage was increased the

TIC increased until about –3400Vwhere further increases were nolonger observed, but instead a slightdecrease was observed. The signalof AMP also increased withincreasing spray voltage, but itreached a maximum at –2700V andthen sharply declined until no signalwas observed for voltages higherthan –4100V. This data most likelyindicates that above –2700V stableelectrospray is starting to changemore to corona discharge ionization.At voltages higher than –4100V theionization process is most likelyentirely corona discharge.

G.R. Asbury and H.H. Hill: „Negative ion electrospray ...”, IJIMS 2(1999)1, 1-8, p. 5

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 4:shows the resolving power measured for AMP atdifferent temperatures using 100% methanol at5µL/min.

60

70

80

90

100

110

120

130

140

150

Res

olvi

ng P

ower

90 140 190 240Temperature (C)

Figure 4

Figure 5Graph A shows the effect of spray voltage on the TICand graph B shows the effect of spray voltage on themeasured signal intensity for AMP for 100% methanolsolutions. All data was collected at 250 °C and asolvent flow rate of 5µL/min.

200

400

600

800

1000

1200

1400

TIC

(pA

)

Figure 5

-4700 -4200 -3700 -3200 -2700 -2200 -1700

Spray Voltage

B

Spray Voltage

-4700 -4200 -3700 -3200 -2700 -2200 -1700

0

100

200

300

400

500

600

Sig

nal I

nten

sity

(A

MP

)

A

Page 10: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

Solvent, drift gas, and cooling gasflow ratesEach of these flow rates was probedto determine their effect on TIC andsignal intensity of AMP usingmethanol solvent at a spray voltageof –2700V and 250 °C. The flowrates of the drift gas(200-1200mL/min) and the coolinggas (0-1400ml/min) had very littleeffect on the TIC or the AMP signal.No more than a 5% change (data notshown) was observed for the entireranges investigated. The solventflow rate (0-100µL/min ) had moresignificant effects. With no solventflowing no signal was observed.This probably indicates that nocorona discharge exists becausewith a corona ions should beobserved even in the absence ofsolvent flow. A maximum value forboth the TIC and AMP signal wasobserved at 4µL/min. Above4µL/min a slow decrease in the TICwas observed with a 25% decrease(data not shown) in the TIC and theAMP signal at flow rates of100µL/min. The effect on the actualspectra however was more severe.Figure 6 (A and B) shows themobility spectra of AMP at flow ratesof 100µL/min and 4µL/minrespectively. The 100µL/minspectrum has a weaker AMP signal

G.R. Asbury and H.H. Hill: „Negative ion electrospray ...”, IJIMS 2(1999)1, 1-8, p. 6

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 6:The top spectrum (A) shows the mobility spectrum ofAMP at a solvent flow rate of 100µL/min. The peakslabeled * are probably a result of large solvent clusters.The bottom spectrum (B) shows the same compound at4µL/min. All data was taken at 250 °C and a sprayvoltage of –2700V using 100% methanol.

AM

P

**

0.0 5.0 10.0 15.0 20.0 25.0 30.0

Drift Time (msec)

0

100

200

300

400

500

600

700

Sig

nal I

nten

sity

(A

MP

)

AM

P0.0 5.0 10.0 15.0 20.0 25.0 30.0

Drift Time (msec)

0

100

200

300

400

500

600

700

Sig

nal I

nten

sity

(A

MP

)

A

B

Figure 6

Figure 7:shows the AMP signalintensity measured withdifferent organic modifiersat different % compositions.The remaining portion ofeach solvent was water inall cases. The data wastaken at a temperature of250 °C and a flow rate of5µL/min. All of the datapoints were taken at thespray voltage that providedthe maximum signal.

MeOH

ACN

EtOH

IpOH

Modifier (%)

4555657585950

100

200

300

400

500

600

700

Sig

nal I

nten

sity

(A

MP

)

Figure 7

Page 11: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

and a much noisier baseline. In addition twoother unidentified peaks appear in thespectrum. Since the mass spectrometer wasset to exclude ions below m/z 200 these peaksare most likely very large solvent ion clustersresulting from incomplete desolvation at theseextremely high flow rates. Though thespectrum at high flow rates is considerablydegraded relative to the low flow rate spectrumit is encouraging that reasonable signal wasstill obtained. This may indicate that ESI-IMSis capable of handling higher flow rates thanESI-MS. This may allow it to interface totraditional HPLC columns without thenecessary splitting required for massspectrometry.

Organics other than methanol100% methanol solutions showed very goodspray results, however methanol tended to bevery intolerant to water. In fact at 50% waterno signal was seen for AMP. Other solventswere used to see how they handled increasedwater. These solvents were ethanol,2-propanol, and acetonitrile. The signalintensity of AMP at different spray voltageswas measured for each solvent at five differentcompositions (90%-50% in 10% increments).A graphical presentation of each is presentedin figure 7. The points on the figure are thosetaken at the voltage that gave the maximumsignal (all points are not at the same sprayvoltage). In all cases as the % of water wasincreased the signal intensity of AMPdecreased. The rate of decrease was lesssevere for ethanol and 2-propanol than for

G.R. Asbury and H.H. Hill: „Negative ion electrospray ...”, IJIMS 2(1999)1, 1-8, p. 7

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Table 1:Minimum required voltage to induce signal and voltage that provides the highest signal for AMP in different solvents on our atmospheric pressure electrospray ionization ion mobility spectrometer.

78-4700-450050% acetonitrile96-4100-350060% acetonitrile

347-3700-310070% acetonitrile355-3500-290080% acetonitrile411-2900-230090% acetonitrile454-4300-350050% 2-propanol483-3300-270060% 2-propanol581-3100-230070% 2-propanol583-2900-230080% 2-propanol627-2700-210090% 2-propanol389-4700-390050% ethanol403-3300-330060% ethanol492-3100-290070% ethanol545-2700-270080% ethanol550-2500-230090% ethanol

no signalno signalno signal50% methanol130-4900-490060 % methanol256-4500-370070% methanol533-3300-270080% methanol573-2700-250090% methanol

Maximum Signal(arbitrary units)

Voltage ofMaximum Signal

Minimum VoltageSolvent1

1 in all cases the remaining portion is water (v/v).

Page 12: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

methanol and acetonitrile. This was probablydue to the decreased surface tension of thedroplets for the longer chain alcohols. Theresults of these studies are further summarizedin table 1. Table 1 contains the minimum sprayvoltage needed to detect ions, the voltage ofmaximum signal and the maximum signalobserved. In all cases the solutions of2-propanol required the least voltage and themethanol solutions required the most voltage.This was once again probably a result ofreduced surface tension for these solutions.Similar trends are also seen for the voltage atwhich the highest signal was observed. Theseresults point very favorably to using 2-propanolinstead of methanol when it is necessary tospray solutions containing significant amountsof water, however 100% methanol still providedthe best overall results.

ConclusionsNegative ion electrospray ionization is lesssensitive and more influenced by water than ispositive mode electrospray ionization foratmospheric pressure ion mobilityspectrometry. In fact when using methanol nosignal is observed for AMP when more than40% of the solution is water. High waterconcentrations quickly lead to coronadischarge ionization instead of electrosprayionization. Increasing the temperaturegenerally provides better signal both in termsof TIC and AMP intensity. The resolvingpower, however reaches a maximum at about150 °C. High spray voltages also lead tocorona discharges. The maximum intensity ofAMP in pure methanol solutions was seen at–2700V. At voltages above –4100V no signalwas seen for AMP in pure methanol. The driftgas and the cooling gas flow rates have littleaffect on the electrospray process at the flowrates investigated. A solvent flow rate of4µL/min provided the best electrosprayperformance, however signal was seen as highas 100µL/min. In addition to methanol otherorganic solvents can be used for ESI-IMS. Ofthese 2-propanol showed the best resultsespecially for solutions with high waterconcentrations, however 100% methanol stillprovided the best electrospray performance.

AcknowledgementsThe authors would like to thank the U.S. Armyfor Grant DAAG559810107 for partial fundingof this work.Additional support was provided ofAnalytical Chemistry of Pittsburgh for an ACSsummer fellowship to G. Reid Asbury.

References[1] Gieniec, M.L.; Cox, J.,Jr.; Teer, D.; Dole, M. 20th

Annual Conference on Mass Spectrometry and AlliedTopics, Dallas, TX, June 4-9, 1972.

[2] Gieniec, J.; Mack, L.L.; Nakamae, K.; Gupta, C.;Kumar, V.; Dole, M. Biomed. Mass Spectrom. 1984,11, 259.

[3] Smith, R.D.; Loo, J.A.; Ogorzalek, R.R.; Busman, M.Mass Spectrom. Rev. 1991, 10, 359-451.

[4] Fenn, J.B.; Mann, M.; Wong, S.F.; Whitehouse, C.M.Science, 1989, 246, 64.

[5] Schumate, C.B. An ElectrosprayIonization/Nebulization Interface for LiquidIntroduction into an Ion Mobility Spectrometer Ph.D.Thesis, Washington State University, 1989.Schumate, C.B.; Hill, H.H., Jr. Anal. Chem. 1989, 61,601-606.

[6] (a) Wittmer, D.; Chen, Y.H.; Luckenbill, B.K.; Hill,H.H., Jr. Anal. Chem. 1994, 66, 2348-2354. (b)Chen, Y.H.; Hill, H.H., Jr.: Wittmer, D. Int. J. MassSpectrom. Ion Proc. 1996, 154. 1-13.

[7] Shelimov, K.; Jarrold, M.F. J. Am. Soc. MassSpectrom. 1996, 118, 1031. Valentine, S.J.;Anderson, J.G.; Ellington, A.D.; Clemmer, D.E. J.Phys. Chem. B 1997, 101, 3891-3900.

[8] Guevremont, R.; Siu, K.W. M.; Wang, J.; Ding, L.Anal. Chem. 1997, 69, 3959.

[9] Wu, C. Siems, W.F., Asbury, G.R.; Hill, H.H., Jr.Anal. Chem. 1998, 70, 4929-4938.

[10] Chen, Y.H.; Hill, H.H., Jr.; Wittmer, D. J.Microcolumn Sep. 1994, 6, 515-524.

[11] Yamashita, M.; Fenn, J.B. J. Phys. Chem. 1984, 88,4671-4675. Bruins, A.P.; Covey, T.R.; Henion, J.D.Anal. Chem. 1987, 59, 2642-2646.

[12] Straub, R.F.; Voyksner, R.D. J. Amer. Soc. MassSpectrom. 1993, 4, 578-587. Cole, R.B.; Harrata,A.K. J. Amer. Soc. Mass Spectrom. 1993, 4,546-556.

[13] Ikonomou, M.G.; Blades, A.I.; Kebarle, P.J. J. Amer.Soc. Mass Spectrom 1991, 2, 497-505.

[14] Cole, R.B.; Harrata, A.K. Rapid Comm. MassSpectrom. 1992, 6, 536-539. Hiraoka, K.; Kudaka, I.Rapid Comm. Mass Spectrom. 1992, 6, 265-268.

G.R. Asbury and H.H. Hill: „Negative ion electrospray ...”, IJIMS 2(1999)1, 1-8, p. 8

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Page 13: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

AbstractCorona discharge ion mobility spectrometrypermits the determination of n-alkanes andbranched chain alkanes within the low ppmrange. For n-alkanes, one or two product ionpeaks are observed. Two product ion peaksare detectable in lower concentration rangeonly (<30-45 ppm in dependence on physicaland chemical properties of n-alkanes). Withincreasing concentrations, the ion mobilityspectra show one product ion peak with asharp peak profile. The reduced mobilities ofthese peaks decrease regularly with increasingmolecular weight. Series of product ion peaksare obtained for branched chain alkanes.Characteristic fragment ions are additionallyformed in comparison with n-alkanes.

IntroductionCorona discharge ion mobility spectrometry(CD-IMS) permits the detection of compounds,which are hardly to detect by means of theconventional IMS using 63Ni ionization. Thisadvantage of CD-IMS results from thecombination of different ionization processes.The positive product ions are mainly formedthrough proton transfer reactions. The courseof these reactions depends on the protonaffinities and gas phase basicities ofinvestigated compounds as well as on thetemperature and composition of carrier gas[1-4]. These processes are comparable withthose of 63Ni ionization. Additionally, electronimpact and photoionization processes withsubsequent ion-molecule reactions and chargetransfer reactions can be expected for CD-IMS[5].

Therefore, investigations of alkanes by usingspectrometers equipped with 63Ni ionizationsources are limited by low sensitivity due to

their low proton affinities. However, a rapidmethod of determination for these compoundsis desirable because of their significance inenvironmental and industrial analysis as wellas for the analysis of volatile petrochemicalfuels. IMS is generally suitable as field monitormethod due to the weight, size and requiredpower supply of spectrometers. Thus, westudied the ability of CD-IMS to provide definitespectra for alkanes and examined thedetectable range of gas phase concentration.

For the above mentioned reasons, only fewexperimental IMS data are available foralkanes [6-9]. These data result frommeasurements in positive mode usingspectrometers equipped with 63Ni ionizationsources. N-alkyl halides [6] and n-alkanes(C5H12-C15H32) were investigated by Karasek etal. [7]. The ion mobility spectra of alkyl halides[RX] were attributed to formed alkylbackbones. The formation of [R-1]+ productions due to a hydrogen abstraction weresupposed for the most abundant ions. Forn-alkanes, the peaks with lowest reducedmobilities were assigned to [MH]+ and [MNO]+

product ions. In contrast to n-alkyl halides,Karasek et al. [7] observed additional fragmentpeaks for n-alkanes. Bell et al. [9] detectedsharp and symmetric peaks for n-alkanes.These peaks were attributed to [M-1]+ and[M-3]+ ions formed by hydrogen abstraction.

The peak assignment based on a comparisonof ion mobility spectra with spectra obtained byCI-MS and APCI-MS (atmospheric pressurechemical ionization - mass spectrometry).[M-1]+ ions were generally detected usingCI-MS [10]. [M-3]+ and [(M-3)H2O]+ ions wereidentified as the most abundant ions usingAPCI-MS with 63Ni ionization [9].

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Determination of n-alkanes and branched chain alkanes by Coronadischarge ion mobility spectrometry

H. Borsdorf1, H. Schelhorn1, J. Flachowsky1, H.-R. Döring2, J. Stach2

1 Centre for Environmental Research Leipzig-Halle, Department of Analytical Chemistry; PF 2; D-04301 Leipzig, Germany2 Bruker Saxonia Analytik GmbH, Permoserstraße 15; D-04318 Leipzig

Received for review December 14, 1998, Accepted February 24, 1999

Page 14: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

ExperimentalThe positive ion mobility spectra of n-alkanesand branched chain alkanes were recorded.The investigated compounds are summarizedin Tab. 1 and 2. The substances used in thisstudy had a purity of about 99 %. About 300 µlof n-alkanes with a chain length < n-C10 andbranched chain alkanes were filled inpermeation tubes, which consisted of apolyethylene tube with a length of about 35mm, an inside diameter of 4 mm and anoutside diameter of 6 mm. The permeationtubes were placed in sealed head-space vials.For the determination of n-alkanes with a chainlength between n-C10 and n-C19, thehead-space vials were replaced by glasscolumns containing a glass micro capillary withup to 30 µl of substance. The gas streamthrough the permeation vessels was keptconstantly by means of flow controllers (25 l/h).A definite amount of sample gas stream wasdiluted with dried ambient air. This gas stream(25 l/h) was injected into the ion mobilityspectrometer. The concentration of thecompounds in the sample gas stream wascalculated using the weight loss of thepermeation vessels over a period of time byusing a microbalance. This sample introduction

system permits the adjustment ofconcentrations within the low ppm range (1-60ppm; ppm: µg/l(g)). The positive ion mobilityspectra were detected with an ion mobilityspectrometer CD1 (BRUKER). Thisspectrometer is equipped with a membraneinlet. The operational parameters of the ionmobility spectrometer were: temperature ofinlet system: 80 °C; sampling gas flow rate: 25l/h; drift gas flow rate: 25 l/h; electric field:about 245 V/cm; temperature of drift tube: 80°C; pressure: atmospheric pressure. Cleanedair was used as sampling gas and drift gas.

Results and discussionAll investigated n-alkanes provide definitespectra, consisting of up to two peaks. Theintensity ratio between both peaks depends onthe concentration. Fig. 1 depicts a series of ionmobility spectra of n-decane. Theconcentration of n-decane was increased from6 ppm to 45 ppm. Fig. 1 illustrates, that thepeak with lower reduced mobility is detectablein lower concentration range only (< 30-45 ppmin dependence on chain length of n-alkanes).This peak is overlapped at increasingconcentrations. Therefore, a sharp peak profileis observed in higher concentration range. A

H. Borsdorf et al.: „Determination of n-alkanes ...”, IJIMS 2(1999)1, 9-14, p. 10

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 1: CD ion mobility spectra of n-decane in dependence on concentration

K0= 1,54 cm2 / Vs

K0= 1,48 cm2 / Vs

reactant ions(positive)

initial concentration: ≈ 6 ppm

final concentration: ≈ 45 ppm

Series of measurements of n-decane at increasing concentrations

drift time

Page 15: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

stronger dependence on concentration wasestablished for the peak with higher K0 value.This behavior is representative for allinvestigated n-alkanes.

Additional fragment ions are not observed incontrast to the investigations reported byKarasek et al. using ß ionization [7] but inaccordance with the results of investigations byBell et al. [9]. The K0 values for n-alkanes aresummarized in Tab. 1. The reduced mobilitieswere calculated according to the conventionalequation [11]:

Kd

t Ep

Tcm Vs0

2

760273

=

=

** * ( / )

with d = drift length (cm); t = drift time (sec); E= field strength (V/cm); p = pressure (torr) andT = temperature (K).

For the calculation of mass-mobility correlationcurve, the reduced mobilities of the peaks withhigher K0 value were used because of theirstronger dependence on the concentration.Besides, mass-mobility correlation curves werederived from reduced mobility valuesknown from literature in order tocompare the drift behavior of formedproduct ions. The calculatedmass-mobility correlation curves arerepresented in Fig. 2. A linearrelationship was established betweenthe logarithms of molecular weightsand reduced mobilities. The reducedmobility values decrease regularlywith increasing chain length for bothpeaks detected by CD-IMS. Thecalculation of mass differencesbetween both peaks by means of theestablished regression equationprovides values between 15 - 20amu. The comparison of reduced mobilitiesobtained by CD-IMS with thosedetermined by Bell et al. [9] exhibitsnearly identical values for the peakswith lower K0 values (n-hexane1,85/1,86 cm2/Vs; n-heptane1,74/1,76 cm2/Vs; n-octane 1,64/1,67cm2/Vs; n-nonane 1,54/1,58 cm2/Vs;n-dodecane 1,31/1,34 cm2/Vs),although the operational parameters

(ß ionization and air as drift gas at 200 °C)differ considerably. On the other hand, theCD-IMS provides generally lower K0 values forn-alkanes in comparison with the reducedmobilities determined by Karasek et al. [7]using ß ionization at 135 °C. However, aproportionality between the determinedreduced mobilities of both investigations wasestablished (ratio of K0-values 63Ni:CD =1,06±0,008).

Therefore, the mass-mobility correlation curves(Fig. 2) are characterized by very similarshapes. The regression equation for CD-IMS(lg m = -0,52 K0+2,95) shows a similar slopecompared to the corresponding curvesobtained by ß ionization (lg m = -0,51 K0+2,98 /lg m = -0,55 K0+2,96). Considering the above described findings, theformation of product ions with a similar driftbehavior can be assumed for both ionizationmethods. Obviously, the peaks detected byCD-IMS can be assigned to product ionsformed by hydrogen abstraction and clusteringwith water. The formation of water clusters isthe most probable pathway due to theestablished mass differences in the

H. Borsdorf et al.: „Determination of n-alkanes ...”, IJIMS 2(1999)1, 9-14, p. 11

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Kd

t Ep

Tcm Vs0

2

760273

=

=

** * ( / )

Table 1: Reduced mobility values for n-alkanes

1,08n-Octadecane

1,081,10n-Heptadecane

1,131,15n-Hexadecane

1,171,20n-Pentadecane

1,211,26n-Tetradecane

1,291,32n-Tridecane

1,341,38n-Dodecane

1,391,45n-Undecane

1,481,54n-Decane

1,581,62n-Nonane

1,671,73n-Octane

1,761,83n-Heptane

1,861,96n-Hexane

2,09n-Pentane

Reduced mobilities (cm2/Vs)

Substance

Page 16: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

mass-mobility correlation curve. Thisassumption is in agreement with the APCI-MSinvestigations by Bell et al. [9]. [M-3(H2O)]+

product ions were identified for n-alkanes inthis study. Fig. 3 shows calibration curves for selectedn-alkanes. An increasing insensibility can be

H. Borsdorf et al.: „Determination of n-alkanes ...”, IJIMS 2(1999)1, 9-14, p. 12

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 2: Mass-mobility correlation curves for n-alkanes

Comparison of mass - mobility correlation curves for n-alkanes

lg m = -0,52 K0 + 2,95

lg m = -0,51K0 + 2,98

lg m = -0,55 K0 + 2,96

1

1,2

1,4

1,6

1,8

2

2,2

2,4

2,6

2,8

3

0,9 1,1 1,3 1,5 1,7 1,9 2,1 2,3

reduced mobility (cm2 / Vs)

lg m

K0 values (n-C5 - n-C15) determined by IMS using 63Ni ionization(Karasek et al.; Anal.Chem. 46 (1974) 970-973)

K0 values (n-C5 - n-C17) determined by IMS using corona discharge ionization

K0 values (n-C6 - n-C9; n-C12) determined by IMS using 63Ni ionization(Bell et al.; J.Am.Soc.Mass Spectrom. 5 (1994) 177-185)

Figure 3: Response curves for selected n-alkanes

Calibration graphs for selected n-alkanes

0

10

20

30

40

50

60

70

80

0 20 40 60 80 100 120

ppm (µg/l)

Inte

nsity

(pA

)

n-pentane

n-heptane

n-nonane

Page 17: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

observed with increasing chain length ofn-alkanes. The reason for the variation of thesensitivities can be caused by the differentpermeabilities of n-alkanes through themembrane inlet used in the CD-IMS due totheir different chemical and physical properties.The detection limits for n-alkanes vary between1-10 ppm. The measured product ion currentsvaried between 3 and 13 % (standard deviationof 7 determinations).

The determined K0 values of investigatedbranched alkanes are summarized in Tab. 2.Contrary to n-alkanes, series of product ionswere detected. Obviously, fragment ions areadditionally formed. The reduced mobilities ofthese ions correspond to those of shortern-alkanes. The reduced mobility of product ionpeak with the highest drift velocity iscomparable with the K0 value of the product ionpeak of n-alkanes with corresponding carboncontent for most investigated branchedalkanes. Besides, product ions with a lower

H. Borsdorf et al.: „Determination of n-alkanes ...”, IJIMS 2(1999)1, 9-14, p. 13

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Table 2: Reduced mobility values for branched chain alkanes

1,571,842,172,2,5-Trimethyhexane

1,451,631,841,972,152,3,4-Trimethylpentane

1,731,852,2,4-Trimethylpentane

1,711,842,162,5-Dimethylhexane

1,631,741,962,053-Methylheptane

1,852,162,4-Dimethylpentane

1,471,631,731,852,052,152,3-Dimethylpentane

1,721,862,053-Methylhexane

1,521,842,152,3-Dimethylbutane

1,521,842,152,2-Dimethylbutane

1,471,641,882,043-Methylpentane

Reduced mobilities (cm2/Vs)Substance

Figure 4:

Ion mobility spectra of binary mixtures of aliphatic hydrocarbons

Ion mobility spectrum of a binary mixture (15 ppm n-nonane and 16 ppm 2,5-dimethylhexane)

K0= 1,84 cm2/Vs (2,5-dimethylhexane)

K0= 1,62 / 1,58 cm2/Vs (n-nonane)

reactant ions(positive)

drift time

Page 18: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

reduced mobility in comparison to the lowestexpected K0 value were detected for fewbranched alkanes at higher concentrations.Therefore, the formation of adduct ions can besupposed. Considering these results, theformation of product ions is probably initiatedby electron impact processes and subsequentgas-phase reactions. The occurrence offragment ions is typically for electron impactprocesses. On the other hand, theidentification of adduct ions indicate theionization through gas-phase reactions.

Fig. 4 depicts the ion mobility spectrum of abinary mixture of 15 ppm n-nonane and 16ppm 2,5-dimethylhexane. As can be seen fromthis spectrum, the used CD ion mobilityspectrometer permits the distinction of differentaliphatic hydrocarbons. However, theconcentrations must be roughly comparableand the investigated compounds have toexhibit a sufficient mass difference. Thepresented example is representative for theinvestigated compounds. Unfortunately, theoccurrence of branched chain alkanescomplicate the peak assignment due to thenumber of detectable peaks and deterioratedresolution of spectra. For mixtures ofn-alkanes, the spectra are generallysuperimpositions of individual spectraconsisting of the peaks with higher reducedmobility.The CD-IMS provides definite spectra forn-alkanes as well as branched chain alkanes.These compounds are detectable within thelow ppm range. The spectra of n-alkanesconsist of up to two peaks. This restrictednumber of peaks permits the identification ofn-alkanes in mixtures. In contrast to this, seriesof product ion peaks are observed forbranched chain alkanes. This detectability ofthese compounds offers new applications ofIMS in environmental and industrial analysis.

AcknowledgmentsThe authors would like to acknowledge theDeutsche Bundesstiftung Umwelt, Osnabrück,for financial support of this work.

References[1] S. H. Kim, K. R. Bettly, F. W. Karasek: Mobility

behavior and composition of hydrate positivereactant ions in plasma chromatography withnitrogen carrier gas. - Anal. Chem. 50 (1978)2007-2012.

[2] J. Sunner, G. Nicol, P. Kebarle: Factors determiningrelative sensitivity of analytes in positive modeatmospheric pressure ionization mass spectrometry.- Anal. Chem. 60 (1988) 1300-1307.

[3] J. Sunner, M. G. Ikonomou, P. Kebarle: Sensitivityenhancements obtained at temperatures inatmospheric pressure ionization mass spectrometry.- Anal. Chem. 60 (1988) 1309-1313.

[4] Z. Karpas, Z. Berant: Effect of drift gas on mobility ofions. - J. Phys. Chem. 93 (1989) 3021-3025.

[5] J. Adler, G. Arnold, H.-R. Döring, V. Starrock, E.Wülfing: First results with the Bruker Saxonia coronadischarge IMS. - 6th Int. Workshop on Ion MobilitySpectrometry, Dresden, Saxony, 10.-14.8.1997,Abstracts (in press).

[6] F. W. Karasek, O. S. Tatone, D. W. Denny Plasmachromatography of the n-alkyl halides. - J.Chromatogr. 87 (1973) 137-145.

[7] F. W. Karasek, D. D. Denny, E. H. Decker: Plasmachromatography of normal alkanes and itsrelationship to chemical ionization massspectrometry. - Anal. Chem. 46 (1974) 970-973.

[8] D. R. Kojiro, M. J. Cohen, R. M. Stimac, R. F.Wernlund, D. E. Humphry,N. Takeuchi:Determination of C1-C4 alkanes by ion mobilityspectrometry. - Anal. Chem. 63 (1991) 2295-2300.

[9] S. E. Bell, R. G. Ewing, G. A. Eiceman: Atmosphericpressure chemical ionization of alkanes, alkenes andcycloalkanes. - J. Am. Soc. Mass Spectrom. 5(1994) 177-185.

[10] T. Keough: Dimethyl ether as reagent gas for organicfunctional group determination by chemicalionization mass spectrometry. - Anal. Chem. 54(1982) 2540-2547.

[11] G. E. Sprangler: Theory and technique for measuringmobility using ion mobility spectrometer. - Anal.Chem. 65 (1993) 3010.

H. Borsdorf et al.: „Determination of n-alkanes ...”, IJIMS 2(1999)1, 9-14, p. 14

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Page 19: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

AbstractsIn the case of investigations of mixtures ofanalytes using ion mobility spectrometers (IMS)peak overlapping and additional peaks,occurring because of dimers containingdifferent molecules of the analytes, oftenreduce the resolution of the spectra. Therefore,the influence of inter-molecular charge transferreactions on the signal should be reduced bytime delayed sample introduction into theionization regions of the IMS realized bycoupling Multi-Capillary columns (MCC) to theIMS. The aim of this combination is to achievefurther enhancement of the resolution of theinstrument and to significantly increase thescope of application of ion mobilityspectrometry. The response of an UV - ionmobility spectrometer (IMS) totrans-1,2-dichloroethene, trichloroethene andtetrachloroethene and the advantages of thecombination of multMCC with ion mobilityspectrometers will be discussed.

The advantages of MCC-IMS compared toMCC coupled to a photoionization detectorswill be demonstrated by achieving efficientseparation of mixtures at ambient temperaturewithin about 1 minute. The results arediscussed with respect to the linear range andthe minimum detectable limit of the 10.6 eV -UV - ion mobility spectrometer particularly inregard on field applications.

IntroductionAlthough lots of methods (especially gaschromatography) exist to separate complexmixtures, the disadvantages to these are that

they are neither continuous nor inexpensive1-3.However, especially for field or processanalysis ion mobility spectrometry could deliveron-line investigations by acquiring spectraeach 50 ms at moderate costs and at ambienttemperature as shown later in this paper. Theeffective separation without heating of theMCC and the detection system is also relevantfor field applications.

The determination of volatile organiccompounds (VOC) at trace levels in aqueousand gaseous samples is of considerableimportance in environmental, medical andprocess applications. The interest incontinuous monitoring of halogenated organicshas continued to grow in the last few years.The on-line determination of specific volatileorganic compounds in e.g. wastewater andgroundwater is a considerable analyticalproblem. Since the inception of the ion mobilityspectrometer (IMS) in 1970 4-5 its potentialsuitability for the detection of volatile organiccompounds has been suggested. Theadvantages of an IMS are low detection limits,good sensitivity as well as low cost and, incomparison to other systems, easytransportation. However, the ionization iscommonly achieved using a radioactive source(e.g. 63Ni) generating bureaucratic expenditure.Other disadvantages of the IMS are non-linearresponse, poor selectivity and interactions ofthe reactant ions with contaminatingcompounds of the sample. Therefore,difficulties often occur in the process ofquantification, especially in the case of mixtureanalysis. An alternative ionization source are

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Detection of trans-1,2-dichloroethene, trichloroethene andtetrachloroethene using Multi-Capillary Columns Coupled to Ion

Mobility Spectrometers with UV-Ionisation Sources

S. Sielemann1, J.I. Baumbach1, P. Pilzecker2, G. Walendzik3

1 Institut für Spektrochemie und Angewandte Spektroskopie (ISAS), Bunsen-Kirchhoff Str.11, D-44139 Dortmund, Germany2 G.A.S. Gesellschaft für Analytische Sensorsysteme mbH Emil-Figge-Str. 76-80, D-44227 Dortmund, Germany3 Insitut für Entsorgung und Umwelttechnik (IFEU), Kalkofen 6, D-58638 Iserlohn, Germany

Dedicated to Prof. Dr. D. Klockow on the occasion of his 65th Birthday

Received for review November 12, 1999, Accepted December 13,1999

Page 20: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

UV-lamps, which are routinely used inphotoionization detectors (PID) in gaschromatography 6-7.

To overcome the insufficient resolution andpeak overlapping to achieve an effectivequantification of mixtures a pre separation ofthe analyte should be helpful because of timedelayed sample introduction into the ionizationregion of the IMS. Because of the possibility todrive it at gas flow rates commonly used in ionmobility spectrometry MCC are investigated.for this purpose. The separation efficiency ofthese columns with retention times between afew seconds and a few minutes at ambienttemperature based on ahigh number oftheoretical plates percolumn (about 5000 permeter) is acceptable forlaboratory and fieldanalysis. Due to a crosssection of 3 mm itspossible to operate theMCC at flow rates up to300 mL/min. Across-sectionalphotograph of a MCC isshown in Figure 1 8.MCC’s are commerciallyavailable with differentstationary phases andlength. First results to

characterise MCC’s have beenpublished previously 9.

Experimental method andprocedure

Sample introductionA schematic diagram of themeasurement system is shown inFigure 2. This consists of a unit forsample preparation, the injectionport, a MCC and the 10.6 eV -UV-IMS. For the calibration of thesystem an exponential dilution flask(EDF) was used, which is acalibration technique frequentlyemployed for gas detectors 10-11. Thedesign of the EDF is shown in Figure3. The glass flask, containing

nitrogen 5.0 (99.999%), has a volume of 5.8 L.On the lid of the flask is a union-T (Swagelok,BEST, Dortmund, Germany) connected on oneside (A) with the carrier gas line for dilution andon the second side (B) with the measurementsystem. On the third side (C) it is possible toinject a liquid sample through a septum. Highpurity nitrogen 5.0 (99.999 %) was used as thecarrier gas (dilution gas flow rate 91 mL/min) toavoid impurities in the carrier gas or humidityfrom the laboratory environment to reduce thetransmission rate of the UV-light through theLiF-window of the lamp and the ionisation yieldand to prevent cluster reactions. The gaseoussamples were injected using a six-port-valve

S. Sielemann et al.: „Detection of trans-1,2-dichloroethene ...”, IJIMS 2(1999)1, 15-21, p. 16

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 1: Cross section of the Multi-Capillary column

Figure 2:Schematic presentation of the measuring system

Carrier gas in

Gas out

UV-Lamp

Shutter Grid

Aperture Grid

Faraday Plate Drift gas in

Driftring

Multi-Capillary Column (MCC)

Electrode

6-Port-Vent

Gaseous Standard

Sample in

Sample Preparation Separation Detection

Sample Loop

Waste

Page 21: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

with a 250 µL stainless steel gas tight loop(Valco Instrument Co. Inc., Houston, USA) ontothe MCC, which was inserted directly into theionization region of the IMS. For continuousmeasurements the sample was introduceddirectly into the ionization region of the IMS. Inthis case the EDF was connected with the IMSusing an approx. 50 cm long piece of teflontubing with a diameter of 2 mm (FischerScientific, Schwerte, Germany). Ion mobility spectrometerThe operating principle of an IMS has alreadybeen frequently described 12-20, therefore only afew essentials are outlined here. In general, acontinuous stream of a gas (air ornitrogen) carrying the analytespasses through an ionizationregion (63Ni ß-radiation, UV-light orvarious types of discharges assource). The analyte molecules areionized directly or via ion-moleculereactions with ionized carrier gasmolecules (proton transfer,electron attachment). Ions formedin the ionization region of the IMSare periodically introduced via ashutter grid into the drift tube,where during their drift under theinfluence of an uniform electricfield they collide with neutral

molecules. Through these collisions,ions attain constant drift velocitiesinversely related to their mass, sothat their collection on a Faradayplate at the end of the drift tubedelivers time dependent signalscorresponding to the mobilities of thearriving ions. This ion mobilityspectrum contains information on thenature of the different tracecomponents present in the carriergas. The Institute ofSpectrochemistry and AppliedSpectroscopy custom designed IMShas been equipped with a 10.6 eV -UV-lamp (EG&G, Shannon FreeZone, Ireland). The main parameterof the IMS are summarised in Table1. For comparison of the calculatedsample amount with the area of themeasured signal the shutter grid wasswitched off and the UV-IMS wasdriven in the same way as a

photoionization detector (PID).

Multi-Capillary columnA rod-shaped MCC (Sibertech, Novosibirsk,Russia) with a length of 21.5 cm was used forthe measurements of the single substances.This column is coated with a 5 %phenyl-methylsilicone stationary phase. Theflow rates of the carrier gas nitrogen 5.0(99.999 %) was about 91 mL/min. For theseparation of the mixtures a spiral MCC SE-30(Alltech GmbH, Unterhaching, Germany) with alength of 70 cm was employed. The carrier gas

S. Sielemann et al.: „Detection of trans-1,2-dichloroethene ...”, IJIMS 2(1999)1, 15-21, p. 17

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 3:Exponential dilution flask

Measurement System

Dilution gas (Nitrogen)

Sample Injection

Gas Exit: Nitrogen + Sample

Septum

5L-Glass Flask

A C

B

Table 1:Parameter of the ISAS custom designed 10.6 eV -UV-IMS

Gate width (ms)Gate width (ms)100Pulse width (µs)

Ambient (22-24°C)Temperature500-800Drift gas flow (mL/min)

Nitrogen 5.0 (99.999 %)Drift gas375Drift voltage (V/cm)615Drift length (mm)507Reaction region (mm)

UV (10.6 ev)IonizationParameter

Page 22: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

flow for the separation of the halogenatedalkenes was fixed at 200 mL/min.

ChemicalsTrans-1,2-dichloroethene (DCEE, 95 %);trichloroethene (TCEE, puriss. p.a., 99.5 %)and tetrachloroethene (PCEE, ”infraredspectroscopic grade”) wereused as obtained from Fluka,Deisenhofen, Germany.Nitrogen (5.0, 99.999%) wasdelivered byMesser-Griesheim,Dortmund, Germany.Procedure. First theexponential dilution unit wascalibrated. As an exampletrichloroethene wasmeasured both continuouslyand using a rod-shaped MCCfor comparison. Theseparation power wasincreased using a longerspiral MCC to demonstratethe efficiency of theseparation oftrans-1,2-dichloroethene,trichloroethene andtetrachloroethene.

Results and discussionTo verify the correct exponential dilution thespectra were measured using theMCC-UV-IMS driven as a PID and the signalarea obtained compared with the calculatedsample amount. Figure 4 shows the

exponential curves of the integratedsignal area and the calculatedsample amount for trichloroethene.After comparison it can be seen thatthe value for the calculated sampleamount is about 20 % higher thanthe measured signal area, butalmost 100 minutes after theinjection a good correspondence tothe measured integral of the signalarea occurs. This delay is due tonon-equilibrium of dilution andsample distribution within the bottle.Therefore, in the followingexperiments with IMS measurementswere taken after 100 minutes hadpassed. The ion mobility spectra oftrichloroethene, which was

introduced continuously, were obtained in aconcentration range of 3 orders of magnitudebetween 25 ppbv and 45 ppmv. Figure 5 showssingle spectra at various concentrations. Theexponential dilution curve (peak height) isshown on the right hand side. The value of 25

ppbv is not the limit of detection as shown inthe single spectra presented on the left handside of Figure 5 and lower concentrations areattainable.

In the case of the continuous introduction oftrichloroethene the signal areas of the peakswere used for calibration and the curve in

S. Sielemann et al.: „Detection of trans-1,2-dichloroethene ...”, IJIMS 2(1999)1, 15-21, p. 18

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 4:Comparison of the calculated sample amount with theintegrated signal area (measured with the UV-IMSdriven as PID)

0 100 200 300 400

Inte

grat

ed S

igna

l Are

a

Injection

Measured Signal Area Calculated Sample Amount

Time / min

Figure 5:Single Spectra for trichloroethene in a concentration rangebetween 25 ppbv and 45 ppmv, insert: single spectrum oftrichloroethene of 24,7 ppbv

0 10 20

05101520

Sig

nal /

a.u

.

Exponential Dilution

24,7 ppbv

24,7 63,2

221,2

905,8 ppbV

4,69

11,1

20,8

38,845,5 ppm

V

Drift time / ms

Page 23: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

Figure 6 was obtained. Thecorrelation is rather high with acorrelation factor of about 0.987.

The UV-IMS was coupled to the 21.5cm rod-shaped MCC and calibratedagain for trichloroethene. Figure 7shows the calibration curve with alinear range between 6 ng and 537ng. The figure also shows the ionmobility spectra at various sampleamounts (5.7 / 40.5 / 537 ng). Thus,for single component measurements,a rod-shaped MCC is applicable. Inthese experiments the concentrationwas varied over 2 orders ofmagnitude.

For the fast separation of mixtures(trans-1,2-dichloroethene, trichloro-ethene and tetrachloroethene) a 70cm spiral SE-30 was used toenhance the separation efficiency. Inthis case the separation was realisedwithin 40 s. Figure 8 shows the3D-plot and the IMS-chromatogram.The single ion mobility spectra arebroad and it is clear, that without preseparation the peaks will overlap inthe drift time region etween 7 msand 10 ms. However, the columnused achieved the full separation. A3-dimensional integration would givethe peak area in this case.

S. Sielemann et al.: „Detection of trans-1,2-dichloroethene ...”, IJIMS 2(1999)1, 15-21, p. 19

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 7:Calibration curve for the non-continuous introduction oftrichloroethene (MCC-UV-IMS)

40.5 ng

2

10 100 1000

10-1

100

101

102

103

Trichloroethene

Sig

nal A

rea

/ a.u

.

Sample Amout / ng

537 ng

5.7 ng

Figure 6:Calibration curve for the continuous introduction of trichloroethene (UV-IMS)

10 100 1000 10000 10000010-2

10-1

100S

igna

l Are

a / a

.u.

Concentration / ppbv

Figure 8:3D-Plot (left) and IMS-Chromatogramm (right) of the mixture of trans-1,2-dichloroethene,trichloroethene and tetrachloroethene (MCC-UV-IMS)

0 20 40 60

8

10

120,05

0,10

0,15

Sig

nal /

a.u

.

Drift Time / m

s

Retention Time / s

0 20 40 60

8

10

12

Tetrachloroethene

Trichloroethene

1,2-Dichloroethene

Drif

t Tim

e / m

s

Retention Time / s

Page 24: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

It is interesting to compare the results from theinstruments driven as a photoionisationdetector and as a ion mobility spectrometer.Figure 9 depicts the calibration curves whenthe spectrometer is operating as a PID(sample amount between 100 pg and 2 µg).The results using the equipment as an IMS areshown in Figure 10 with sample amountsbetween 30 ng and 5.1 µg. These resultsdemonstrate clearly that the PID is moresensitive than the IMS. This is a result of thefact that the shutter of an IMS opensperiodically for 1 ms and only the ions driftingthrough the shutter during this time interval canreach the Faraday plate. In the PID the shutteris switched off during the whole time and allions can reach the electrode. However, anadvantage of the IMS is that drift time andretention time can be used to identify thepeaks of the signals as shown recently in thecase of investigations of contaminatedindustrial wastewater21.

The detection limits fortrans-1,2-dichloroethene, trichloroethene and

tetrachloroethene are in the range of 31 ng, 36ng and 59 ng, respectively. For singlesubstances (see Figure 7) the minimumdetectable sample amount of trichloroethenewas about 6 ng. This is lower than for mixtureanalysis because of the lower flow rates ofcarrier gas and the drift gas flow. For theseparation of the mixtures its necessary towork with higher flow rates to reach a higherresolution of the signals in the chromatogram.

ConclusionsAn ISAS custom designed IMS with a 10.6 eV -UV-ionization source was used for thedetection of trichloroethene, which wascontinuously introduced. A detection limit ofless than 25 ppbv was achieved. To measuremixtures the coupling with a MCC leads to afast separation of 1,2-dichloroethene,trichloroethene and tetrachloroethene within atime interval of 40 s with a linear rangebetween 30 ng and 5.1 µg. Calibration usingan exponential dilution flask leads to excellentagreement between calculated amount andmeasured signal area.

S. Sielemann et al.: „Detection of trans-1,2-dichloroethene ...”, IJIMS 2(1999)1, 15-21, p. 20

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 10: Calibration curves fortrans-1,2-dichloroethene, trichloroethene and tetrachloroethene (UV-IMS-MCC)

10 100 1000 10000

0,1

1

Sample Amount / ng

Tetrachloroethene

0,1

1

Trichloroethene

Sig

nal A

rea

/ a.u

.

0,1

1

1,2-Dichloroethene

Figure 9:Calibration curves fortrans-1,2-dichloroethene, trichloroethene andtetrachloroethene (UV-IMS-MCC driven asPID)

0,01 1 100 10000

0,01

1

Sample Amount / ng

Tetrachloroethene

0,01

1

Trichloroethene

Sig

nal A

rea

/ a.u

.

0,01

1

trans-1,2-Dichloroethene

Page 25: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

When the IMS is used as a photoionisationdetector the linear range is about one order ofmagnitude higher and the detection limit ingeneral lower. Nevertheless, the use of MCCwith an IMS using UV-ionisation sources leadsto an increase in the scope of application by areduction in overlapping peaks. Especially inthe case of quantitative investigations ofmixtures of different gases the possibility torealize effective resolution is enhanced

AcknowledgmentsA part of the results presented here wereobtained with the financial support of the European Union (contract EV5V-CT94-0546).The financial support of the Bundesministeriumfür Bildung, Wissenschaft, Forschung undTechnologie and the Ministerium fürWissenschaft und Forschung des LandesNordrhein-Westfalen and the co-operation withthe Institute of Applied Physics, the Institute ofCatalysis and the company Sibertech,Novosibirsk, Russia, are gratefullyacknowledged.

References [1] B.J. Harland, P.J.D. Nicholson, E. Gillings,

Determination of volatile organic compounds inaqueous systems by membrane inlet massspectrometry, Water Research, 21, 1, 107-113,1987.

[2] G. Eklund, B. Josefsson, C. Roos, Determination ofvolatile halogenated hydrocarbons in tap water,seawater and industrial effluents by glass capillarygas chromatography and electron capture detection,J. High Resol. Chromatograph. Chromatograph.Commun., 1, 34-40, 1978.

[3] C.J. Koester, R.E. Clement, Analysis of drinkingwater for trace organics, Crit. Rev. Anal. Chem., 24(4), 263-316. 1993.

[4] M.J. Cohen, F.W. Karasek, Plasma chromatography– a new dimension for gas chromatography andmass spectrometry, J. Chromatogr. Sci., 18, 88-92,1970.

[5] Z. Karpas, Y.-F. Wang, G.A. Eiceman, Qualitativeand quantitative response characteristic of a capillarygas chromatograph/ion mobility spectrometer tohalogenated compounds, Anal. Chim. Acta, 282,19-31,1993.

[6] M.A. Baim, R.L. Eatherton, H.H. Hill, Ion mobilitydetector for gas chromatography with a directphotoionization source, Anal. Chem., 55, 1761-1766,1983.

[7] J.W. Leonard, W. Rohrbeck, H. Bensch, A highresolution IMS for environmental studies, FourthInternational Workshop on Ion MobilitySpectrometry, 16.8.-9.8.1995, Cambridge, U.K.

[8] Institute for Catalysis and Company Sibertech,Novosibirsk, RUS, private Information.

[9] J.I. Baumbach, G.A. Eiceman, D. Klockow, S.Sielemann, A. v.Irmer, Exploration of amulticapillary column for use in elevated speedchromatography, Int. J. Env. Anal. Chem. 66,189-193, 1997

[10] J.J. Ritter, N.K. Adams, Exponential dilution as acalibration technique, Anal. Chem., 48 (3), 612-619,1976.

[11] J.M. Sedlak, K.F. Burton, Comments on use ofexponential dilution flask in calibration of gasanalysers, Anal. Chem., 48 (13), 2020-2022, 1976.

[12] H.E. Revercomb, E.A. Mason, “Theory of plasmachromatography/ gaseous electrophoresis - a review”Anal. Chem. 48, 970-983, (1975).

[13] E.W. McDaniel, E.A. Mason, “The mobility anddiffusion of ions in gases”, John Wiley, New York,(1973).

[14] T. Carr, “Plasma Chromatography”, Plenum, NewYork, (1984).

[15] H.H. Hill, W.F. Siems, R.H. St.Louis, D.G. McKinn,“Ion mobility spectrometry” Anal.Chem. 62,1201A-1209A, (1990).

[16] R.H. St.Louis, H.H. Hill, “Ion Mobility Spectrometry inAnalytical Chemistry” Crit. Rev. Anal. Chem. 21,321-355, (1990).

[17] H.M. Widmer, M.A. Morrissey, “Neochromatographictechnologies - ion mobility spectrometry” Chimia 43,268-277, (1989).

[18] J.E. Roehl, “Environmental and Process Applicationsfor Ion Mobility Spectrometry” Appl. Spectrosc. Rev.26, 1-57, (1991).

[19] G.A. Eiceman, “Advances in Ion MobilitySpectrometry” Crit. Rev. Anal. Chem. 22, 17-36,(1991).

[20] J.I. Baumbach, G.A. Eiceman, "Ion mobilityspectrometry: Arriving on-site and moving beyond alow profile" Applied Spectroscopy 53, 338A-355A,(1999)

[21] A.N. Davies, J.I. Baumbach: Multidimensional dataanalysis - quantifying the hidden dimension. - Spectroscopy Europe 11, 23-24, (1999)

S. Sielemann et al.: „Detection of trans-1,2-dichloroethene ...”, IJIMS 2(1999)1, 15-21, p. 21

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Page 26: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

AbstractIon mobility spectrometry (IMS) has beenknown as an analytical technique since the late1960s and early 1970s. To date, it has beensuccessfully utilized for the detection ofenvironmental pollutants, warfare agents,explosives, herbicides, pesticides, petroleumproducts as well as for the detection ofprescription and illicit drugs. In this paper theauthors describe the first IMS field tests inGermany and Austria. We also report the useof the IMS methodology for the rapid analysisof hallucinogenic fungi material. A newapplication of the IMS technology for theanalysis of postmortem sweat samples fordrugs is also presented.

IntroductionAs a new method of analysis ion mobilityspectrometry (IMS) was first introduced byCohen and Karasek in 1970 [1]. Since then thepotential of IMS as an analytical instrument, asensitive trace detector and sophisticatedmonitor is realized. Since the beginning of the1990s the IMS methodology is gaining moreand more importance and acceptance amongenvironmental authorities, military institutionsas well as in the field of forensic sciences.Discussing industrial, environmental or militaryapplications of IMS one has to distinguishbetween fixed-point instruments and hand-heldmonitors. Fixed-point instruments monitor aspecific compound or a number of hazardoussubstances at a fixed location continuously.Hand-held monitors on the other hand aremainly used to detect specific compounds for ashort peroid of time at a certain spot. Sincethese instruments can be transported to anylocation easily they are ideally suitable fordetecting leaks or accumulations of specialchemicals considered hazardous to manand/or to the environment. Ion mobility

spectrometers developed for the rapiddetection of hidden contraband materials(drugs and explosives) are somehow biggerand heavier compared to hand-heldinstruments. Nevertheless they are alsoportable. Ion mobility spectrometry is forexample used for the direct analysis ofexplosives [2-11], atmospheric and workplacepollutants [12-16], warefare agents [17,18] andtear gases [19,20] but also for the analysis ofprescription and illicit drugs [21-28]. In forensicsciences the IMS methodology is mainly usedas a detection device to prove the prescenceof latent traces of illict drugs or explosives onsurfaces of suspicious or confiscated materialof evidence. Due to the fact that drugs andexplosives have only very low vapourpressures efficient sampling techniques arerequired to successfully detect the compoundsof interest. Therefore drug residues are"collected" either by wiping the correspondingobjects with specially treated filters, Q-tips,paper or even tooth picks. The entrapment ofthe drug microparticles is also achieved by theutilization of a specially constructed vacuumcleaner. The particles are directly sucked on aTeflon membrane filter and evaporated at anelevated temperature in a furnace directly intothe IMS system.

Principle of IMSIon mobility spectrometry refers to theprinciples, practice and instrumentation forcharacterizing chemical substances throughtheir gas-phase ion mobilities. IMS is ananalytical technique that distinguishes ionicspecies on the basis of the differences in thedrift velocity through a gas under an appliedelectrostatic field. It is a sensitive technique forthe detection of trace organics underatmospheric pressure conditions. Fig. 1 depictsthe schematic representation of the IONSCAN

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Ion Mobility Spectrometry for the Detection of Drugs in Casesof Forensic and Criminalistic Relevance

Thomas Keller1*, Andrea Schneider1, Edith Tutsch-Bauer1, Jürgen Jaspers2, RolfAderjan3 and Gisela Skopp3

1 Institute of Forensic Medicine, University of Salzburg, Ignaz-Harrer-Str. 79, 5020 Salzburg, Austria2 Telerob, Gesellschaft für Fernhantierungstechnik mbH, Vogelsang-Str. 8, 73760 Ostfildern, Germany3 Institute of Forensic and Traffic Medicine, University of Heidelberg, Voss-Str. 2, 69115 Heidelberg, Germany

Received for review February 25, 1999, Accepted March 15, 1999

Page 27: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

detection system. The sample, collected forexample on a membrane filter, is heated tovaporization by the desorber. The neutralmolecules of the vapor are carried in a streamof dried, filtered, ambient air through theheated transfer line into the reaction region.Here, ionization is initiated by high energyelectrons emitted from a 63Ni beta-ray source.The product ions (positive or negative) aregated with a pulse width of 0.02 ms into theheated drift region for mobility analysis every20 ms. Under the influence of a controlledelectric field and against a counterflow ofambient air drift gas, the ions move to thecollector electrode. The drift times required bythe ions to reach the collector electrode areproportional to their masses but inverselyproportional to their characteristic reduced ionmobilities (K0). The ion mobility spectrometer isoperated in the positive mode for drugdetection. In this mode, the drift gas containstrace amounts of nicotinamide (NTA) used asboth calibrant and reactant. In the reactionregion, the protonated nicotinamide transfers aproton to the sample molecule according toEquation I:

[NTA]H+ + M → NTA + [M]H+ (I)

This reaction onlyproceeds if theproton affinity ofthe samplemolecule is greaterthen that ofnicotinamide. Thisis true for nearly alldrug moleculeswhich are thusdetectable by IMS.The principles andtheoreticalbackground of ionmobilityspectrometry hasbeen extensivelydescribedelsewhere[12,15,29-33] andwill not bediscussed in thispaper.

Ion mobility spectrometry in Heidelberg,Baden-Württemberg, Germany

Police controlCase history and IMS analysis

The first IMS field test in Germany wasconducted in Heidelberg, Baden-Württemberg,in 1997. The tests were performed bymembers of the Institute of Forensic andTraffic Medicine in cooperation with HeidelbergPolice Department and Heidelberg Customs. Inthe tests an ion mobility spectrometer(IONSCAN 400) was used as a detectiondevice in the routine control of suspectedpersons and/or vehicles for hidden drugs ordrug residues. The check point set up byofficers of Heidelberg Police Department waslocated near a popular disco. Visitors on theirway to the disco were checked after theofficers had conducted them to the checkpoint. In the course of the control a car wasstopped and the driver was searched for drugs.During the search a white powder wasconfiscated and analyzed by IMS. Therefore atooth pick was slightly dipped into thecrystalline white powder. Crystals entrapped onthe tip of the tooth pick were then wiped on theTeflon membrane filter and inserted directly

Th. Keller et al.: „Ion mobility spectrometry ...”, IJIMS 2(1999)1, 22-34, p. 23

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 1:Schematic representation of the IONSCAN 400 detection system

Page 28: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

into the IMS system. The analysis proved thatthe white substance was cocaine. The driversclothes and hands were also screened fordrugs. Prior to IMS analysis his clothes werevacuum cleaned. His hands were wiped with aQ-tip slightly moistened with methanol to betterremove the drug residues. Then the sampleswere analyzed by IMS. Hands and clothes alsoproved cocaine positive. The car interior wasalso heavily contaminated by cocaine.Interrogating the driver for his consuminghabits he admitted of having sniffed about 0.3g cocaine on the day of control. Nose smearwas also taken with a Q-tip and analyzed byIMS. The plasmagram of the nose smear isshown in Fig. 2.

Customs controlCase history and IMS analysisThe check point chosen by officers ofHeidelberg Customs was a resting place on aGerman highway known to be used as drugtransportation route. The suspected car driverswere guided to the resting place where thepersons and the cars were inspected. Thesuspected vehicle for example was searchedfor drug residues by wiping the instrumentboard, door handle, gear shift etc. with a cottonswap. The front seats and rear seats werevacuum cleaned. The collected microparticleswere then directly analyzed by IMS. It could beshown that the interior of one car wascompletely contaminated by the designer drug3,4-methylenedioxyethylamphetamine (MDEA),known as "Eve". As soon as it became obviousthat traces of MDEA could be detected insidethe car, the vehicle was also searched fordrugs by customs officers and their trackerdogs. The owner of the car was also searchedfor drugs or drug residues. Therefore hisclothes (shirt and pants), his money, purse, hippocket and hands (including fingernail dirt)were examined. His hands were wiped with aQ-tip slightly moistened with methanol. Allother items were vacuum cleaned. IMSanalysis proved MDEA positive for all itemsanalyzed. During the search of the vehicle 8.5rhombic shaped tablets could be confiscated.Designer tablets of a rhombic shape are prettyrare. The confiscated tablets measured 12.6mm in lenth, 7.4 mm in width and were 4.6 mmthick with a total weight of 306.5 mg. Thetablets and the plasmagram are depicted inFig. 3. Prior to IMS analysis the tip of a tooth

pick was rubbed along the tablet´s surface toremove some crystals. These were thentransfered on the Teflon membrane filter andanalyzed by IMS.

Samples preparation and GC/MSmeasurementAn accurately weighed amount of thepowdered tablet was dissolved in methanol toobtain a final concentration of 1 mg MDEA/mL.The solution was then shaken for 15 min.Centrifugation at 14000 rpm for 10 minresulted in a clear supernatant. One µL of thesupernatant was directly used for GC/MSanalysis. GC was performed on an HP 6890Series gas chromatograph equipped with anHP 5973 mass selective detector. TheTMS-derivatives were separated on a 30 m x0.25 mm i.d. HP-5 MS fused silica capillarycolumn (95 % dimethyl-5 % diphenylpolysiloxane) with a 0.25 µm film thickness withhelium as the carrier gas at a constant flowrate of 1 mL/min. The splitless mode was usedwith 1 µL samples being injected. Theoperating conditions for the analyses were:injection port temperature 250 °C; initialtemperature 130 °C, programming 20 °C/min tofinal temperature, 270 °C. The MSD wasoperated in the SIM mode using the ions m/z207, 163 and 135. The peak areas of all threemass fragments were used for quantitation.Analyses were performed in duplicate. Thetablets contained 25.4 mg (8.3 %) MDEA.

Ion mobility spectrometry in Salzburg,AustriaCase 1Case history and IMS analysisIn the course of a border patrol about 1 kg ofso far undescribed fungus material wasconfiscated. For IMS analysis the cap andstem of the fungus was simply wiped with atooth pick. The microparticles entrapped on thetip of the tooth pick were then transferred onthe Teflon membrane filter and inserted intothe IMS system. Small pieces of the cap andstem were also used for IMS analysis. Theconfiscated fungi are depicted in Fig. 5. IMSanalysis revealed that both the cap and thestem contained the hallucinogenic compoundspsilocybin and psilocin. The typicalplasmagrams of the mushroom cap and stemof Psilocybe subcubensis are presented inFigs. 9 and 10. The confiscated fruit bodies of

Th. Keller et al.: „Ion mobility spectrometry ...”, IJIMS 2(1999)1, 22-34, p. 24

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Page 29: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

the investigated species were determined asPsilocybe subcubensis Guzmán. The distinctannulus on the stem, the blueing of the wholefruit body and the size of the spores arecharacteristic for this species. As a "negative"sample, fruit bodies of the species Agrocybepraecox (Pers.) Fayod were chosen.Macroscopic characters of Agrocybe praecoxare very similar to Psilocybe subcubensis butAgrocybe praecox does not contain thehallucinogenic components psilocybin orpsilocin. Fig. 8 shows the plasmagram of thenegative control.

Samples, sample preparation and GC/MSmeasurementParts of fungal fruit bodies were lyophilizedand the freeze-dried cap and stem were cutinto small pieces and ground to a fine powderin a mortar. An accurately weighed amount ofthe corresponding powdered sample (50 mg)was extracted with 1 mL chloroform in anultrasonic bath for 1 hour. Centrifugation at14000 rpm for 10 min and filtration through acotton filter resulted in a clear supernatant. 0.5mL of the filtrate from spiked mushroomsamples was pipetted into a GC vial andevaporated to dryness at 50 °C under a gentlestream of nitrogen. The same procedure wasapplied to the psychedelic mushroom materialexcept that only 0.05 mL of the filtrate wasevaporated. Each residue was dissolved in 30µL MSTFA (N-Methly-N-(trimethylsilyl)-2,2,2-trifluoroacetamide) and heated for 30 minat 70 °C. After cooling, one µL of the samplewas directly used for GC/MS analysis. GC wasperformed on an HP 6890 Series gaschromatograph equipped with an HP 5973mass selective detector. The TMS-derivativeswere separated on a 30 m x 0.25 mm i.d. HP-5MS fused silica capillary column (95 %dimethyl-5 % diphenyl polysiloxane) with a0.25 µm film thickness with helium as thecarrier gas at a constant flow rate of 1 mL/min.The splitless mode was used with one µLsamples being injected. The operatingconditions for the analyses were: injection porttemperature 250 °C; initial temperature 180 °C,programming 20 °C/min to final temperature,320 °C for 5 min. The MSD was operated inthe SIM mode using the ions m/z 485, 455 and442 for psilocybin and m/z 348, 291 and 290for psilocin respectively. The last ion listed foreach compound was used for quantitation.

Psilocybin and psilocin were found in anamount of 0.86 % (8.6 mg/g) and 0.02 % (0.2mg/g) per dry mass in the cap of Psilocybesubcubensis, respectively. 0.8 % psilocybin(8.0 mg/g) and 0.03 % psilocin (0.3 mg/g) perdry mass could be found in the stem.Standards of known amounts of psilocybin andpsilocin were prepared by spiking "negative"mushroom samples to obtain finalconcentrations of 0.1, 0.2, 0.9 mg psilocybin/gand 0.01, 0.1, 0.2 mg psilocin/g dry mass.Standard curves for psilocybin and psilocinfitted a linear model over the concentrationrange mentioned. The correlation coefficientscalculated by linear regression analyses were0.997 for both psilocybin and psilocin.

Case 2Case history and IMS analysisIn a case of great public interest an Austrianjudoka (Austrian and European champion) wasarrested while receiving about 1.3 kg cocainefrom his Lithuanian sports colleagues. After thedealers got arrested their clothes and carlicensed in Lithuania was confiscated. Duringthe interrogation the three Lithuanians deniedof having anything to do with the drug.However, investigations of SalzburgGendarmerie in cooperation with the GermanFederal Police Agency (BKA); Interpol Lyon,France; Interpol Lithuanian National Bureauand the Bureau for Organized Crimes Alytus,Lithuania proved that the dealers belonged toa criminal organization named "Tamosial", anorganization of the Russian mafia. In thesearch for a rapid and sensitive method for thedetection of trace amounts of drugs ionmobility spectrometry was applied. The dealersclothes (jackets, shirts, pants, underwear,socks and shoes) and the car interior werethoroughly analyzed by IMS. The air supplyshaft used as cocaine hiding place was alsoexamined. All items under investigation werecocaine positive. Other drugs could not bedetected.

Confirmation analysis by GC/MSEntrapped microparticles analized by IMS wereplaced in a plastic tube and eluted with 2 mLmethanol. Centrifugation at 14000 rpm for 5min resulted in a clear supernatant. Under agentle stream of nitrogen, the organic layerwas evaporated to dryness at 50 °C and theresidue reconstituted into 20 µL methanol. GC

Th. Keller et al.: „Ion mobility spectrometry ...”, IJIMS 2(1999)1, 22-34, p. 25

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Page 30: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

was performed on an HP 6890 Series gaschromatograph equipped with an HP 5973mass selective detector. Cocaine wasseparated on a 30 m x 0.25 mm i.d. HP-5 MSfused silica capillary column (95 % dimethyl-5% diphenyl polysiloxane) with a 0.25 µm filmthickness with helium as the carrier gas at aconstant flow rate of 1 mL/min. The splitlessmode was used with 1 µL samples beinginjected. The operating conditions for theanalyses were: injection port temperature 250°C; initial temperature 220 °C, programming 20°C/min to final temperature, 310 °C for 1 min.The MSD was operated in the SIM mode usingthe ions m/z 182, 198 and 303 for cocaine. Allsamples IMS positive for cocaine could beconfirmed by GC/MS analysis.

Postmortem sweat analysisSamples, sample preparation IMS analysisSpecimens of sweat were obtained fromdeceased drug addicts whose consuminghabits were known. Samples were taken fromthe forehead, breast, arm pit and inguinalregion. Sweat was collected with the cottonpad of a "Salivette" (Sarstedt, Nümbrecht,Germany). Prior to the collection the pad wasslightly moistened with 70 % ethanol. A skinarea of 25 cm2 was then wiped. For IMSanalysis the "sweat pad" was slightly pressedon the Teflon membrane filter. The solvent was

evaporated to dryness and the filter directlyinserted into the IMS system for drug analysis.

Samples, sample preparation GC/MSanalysisFor GC/MS analysis for opiates and cocainethe cotton pad was eluted with 2 mL 0.2 M pH9 sodium carbonate/ sodium bicarbonate buffercontaining 20 ng/mL and 200 ng/mL ofdeuterated analogues as internal standard.Pads and buffer were mixed for 30 min on areciprocating shaker. Centrifugation at 4300rpm for 10 minutes resulted in a clearsupernatant. The organic layer was applied ona Chromabond drug (Macherey&Nagel, Düren,Germany) pre-conditioned C18 extractioncolumn (200 mg/3mL) for further separation.The target drugs were eluted from the columnwith 2 mL dichloromethane-isopropanol 8:2containing 2 % ammonia. The organic layerwas evaporated to dryness and the residuederivatized with 50 µL pentafluoropropionicacid anhydride (PFPA) andhexafluoroisopropanol (HFIP) at 70 °C for 30min. Excess of derivatizing agent wasevaporated under a gentle stream of nitrogenat 50 °C. The derivative extracts were thentaken up in 20 µL dichloromethane/isopropanol9:1. One µL of each sample was analyzed byGC/MS using selective ion monitoring. GC wasperformed on an HP 5890 Series gas

Th. Keller et al.: „Ion mobility spectrometry ...”, IJIMS 2(1999)1, 22-34, p. 26

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 2:Plasmagram of nosesmear

Page 31: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

chromatographequipped with an HP5972 mass selectivedetector. Separationwas performed on a12.5 m x 0.25 mmi.d. CP-Sil 5 fusedsilica capillarycolumn (95 %dimethyl-5 %diphenylpolysiloxane) with a0.4 µm film thicknesswith helium as thecarrier gas at aconstant flow rate of1 mL/min. Thesplitless mode wasused. The operatingconditions for theanalyses were:injection porttemperature 250 °C;initial temperature110 °C for 1 min,programming 20°C/min to finaltemperature, 300 °C.

Results and discussionThe plasmagram of the car driver´s nosesmear is shown in Fig. 2. The drift time ofcocaine was 15.188 ms with a K0

value of 1.1600.

A picture of the rhombic shapeddesigner tablets and thecorresponding plasmagram ispresented in Fig. 3. The drift time ofMDEA was 12.489 ms. The K0 valuewas 1.4216.The confiscated hallucinogenic fungimaterial is presented in Fig. 4.A plasmagram of a psilocybin andpsilocin standard solution inmethanol is shown in Figs. 5 and 6.The drift times for the psychotropicsubstances psilocybin and psilocinwere 11.834 and 11.822 ms,respectively. Due to thermaldephosphorylation of underivatizedpsilocybin in the inlet system of theion mobility spectrometer (or gaschromatograph), psilocybin is

converted into psilocin. The plasmagram of thenegative mushroom sample is shown in Fig. 7.Typical plasmagrams of cap and stem ofPsilocybe subcubensis are depicted in Figs. 9and 10, respectively. The peak detected was

Th. Keller et al.: „Ion mobility spectrometry ...”, IJIMS 2(1999)1, 22-34, p. 27

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 3:MDEA tablet and plasmagram

Figure 4:Dried fruit bodies of Psilocybe subcubensis

Page 32: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

Th. Keller et al.: „Ion mobility spectrometry ...”, IJIMS 2(1999)1, 22-34, p. 28

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 5: Plasmagram of a psilocybin standard solution in methanol (DTime = 11.834 ms(psilocybin))

Figure 6:Plasmagram of a psilocin standard solution in methanol (DTime = 11.822 ms(psilocin))

Page 33: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

Th. Keller et al.: „Ion mobility spectrometry ...”, IJIMS 2(1999)1, 22-34, p. 29

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 8:Plasmagram of caps of Psilocybe subcubensis (DTime = 11.915 ms (psilocybin))

Figure 7:Plasmagram of fruit bodies of Agrocybe praecox (negative control)

Page 34: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

identified as psilocybin on both reduced ionmobility (K0) and drift time (DTime). Throughoutthe IMS experiments, the threshold was set at50 digital units (du). The utilization of Teflonmembrane filters was most effective for theanalysis of the mushroom material. A desorbertemperature of 288 °C (also ideal for the

detection of other illicit drugs) resulted in amaximum peak intensity for psilocybin andpsilocin. Optimized operation parameters aresummarized in Table 1.

The picture of the air supply shaft used ascocaine hiding place is shown in Fig. 10.

Th. Keller et al.: „Ion mobility spectrometry ...”, IJIMS 2(1999)1, 22-34, p. 30

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Table 1: IONSCAN operation parameters for drug analysis

Parameter Setting

Desorber temperature [°C] 288Inlet temperature [°C] 279Drift tube temperature [°C] 235Drift flow [cm3/min] 300Sample flow [cm3/min] 200Stand-by drift flow [cm3/min] 51Gate width [ms] 20Scan period [ms] 20Drift tube length [cm] 7Shutter grid width [ms] 0.2Threshold [du] 50Drift gas dried airCarrier gas dried airCalibrant/Reactant nicotinamide

Figure 9:Plasmagram of stems of Psilocybe subcubensis (DTime = 11.900 ms (psilocybin))

Page 35: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

Cocaine contaminations could be found insidethe shaft as well as on its outer surface.

For many years forensic toxicologists havedetected the presence of illicit drugs andmedications in biological materials mainly usingblood and urine but also body fluids likevitreous humor, cerebrospinal fluid or eventissue samples. In recent years, remarkableadvances in sensitive analytical techniqueshave enabled the analysis of drugs inunconventional biological material such assweat. As early as in 1844 Valentin [32]showed that after the incorporation of chinine,sulphur, copper, indigo or iodine these

substances were excreted intosweat. Benzoic acid could later befound in sweat by Schottin [33].Tachau [34] proved in 1911 thatantipyrine, chinine, benzoic acid,salicylate and alcohol were excretedin sweat after ingestion of thesubstances. "Veronal" was firstdetected by Jansch [35] inconnection with a lethal barbiturateintoxication. Nowadays, the detectionmethods for drugs in sweat includeimmunoassay techniques [36-43]and mostly gas chromatogra-

phy-mass spectrometry [36-50]. Detailedinformation on sample preparation or data onoperation parameters using the IMSmethodology for sweat analysis have not beenpublished in literature so far. The results ofsweat analysis performed by IMS and GC/MSare presented in Table 2.

As demonstrated in Table 2 the resultsobtained by ion mobility spectrometry are ingood agreement with the results obatined byGC/MS. In the cases examined cocaine couldbe detected at different sweat sampling sites.Benzoylecgonine (BE) and ecgoninmethylester(EME) could not be detected by IMS since the

Th. Keller et al.: „Ion mobility spectrometry ...”, IJIMS 2(1999)1, 22-34, p. 31

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 10:Air supply shaft used as cocaine hiding place

Table 2: Sweat analysis using IMS and GC/MS

Cocaine: 64Cocaine: ⊕arm pit4

Cocaine: 141

BE: 477

EME: 12

Cocaine: ⊕forehead3

Cocaine: 41

BE: ⊕

EME: 6

Cocaine: ⊕breast2

Cocaine: 117

BE: 30

EME: 30

Cocaine: ⊕inguinal region1

GC/MS [ng/25 cm2]IMSSampling SiteCase

Page 36: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

spectrometer was not calibrated with thesesubstances. For IMS analysis the thresholdvalue for cocaine was set between 2 and 50digital units (du). 2 du are the minimum valueadjustable thus resulting in a maximum ofsensitivity. Results of the IMS tests areobtained in 8 seconds. So far, when sweatanalysis is performed immunoassaytechniques [36-43] and mostly gaschromatography-mass spectrometry areapplied. The goal of the present study was toindicate that IMS can be a simple and reliabletool for the detection of cocaine in sweat ofdeceased drug addicts. The advantages of ionmobility spectrometry are not only the veryshort analysis time, its detection limits in the ngto pg range but also the possibility to analyzesolid, and pulverized specimens as well assamples in solution (after evaporation of thesolvent). IMS is very selective because thereduced mobilities are intrinsic properties ofindividual ions. In contrast to on-site drug testson an immunological basis [51] false negativeresults are very rare. The false alarm rates ofthe ion mobility spectrometer utilized isreported with 0.3 % in the negative mode andalso less then 1 % in the positive mode [52].The present paper clearly demonstrates thation mobility spectrometry was, for the first time,successfully field-tested in Germany andAustria to rapidly obtain indications of thepresence of illegal drugs on surfaces, fungimaterial, nose smear and sweat. Neverthelessadditional studies on a wider range of drugsand their metabolites are needed to furtherinvestigate the applicability of ion mobilityspectrometry in cases of clinical and forensicrelevance.

References[1] M.J. Cohen and F.W. Karasek.

Plasmachromatography- a new dimension for gaschromatography and mass spectrometry. J.Chromatogr. Sci. (1970) 8: 330-337

[2] D.D. Federrolf and T.D. Clark. Detection of traceexplosive evidence by ion mobility spectrometry.Proc. of the 1. Intern. Symp. on ExplosiveDetection Technology, Atlantic City, N.J., 1991

[3] Th. Keller, A. Chlewinski, R. Binz, P. Regenscheit,and R. Dirnhofer. IMS und HPLC im Kampf gegenSprengstoffattentate. Kriminalistik (1996) 7:513-517

[4] B.J. Yelverton. Analysis of RDX vapors in pre- andpost-detonations using the ion mobility

spectrometer under field conditions. J. Energ. Mat.(1988) 6: 73-80

[5] P. Kolla. Detecting hidden explosives. Anal. Chem.(1995) 67: 184A-189A

[6] L.L. Danylewich-May. Modifications to theionization process to enhance the detection ofexplosives by IMS. Proc. of the 1. Intern. Symp. onExplosive Detection Technology, Atlantic City,N.J., 1991

[7] W. Bernhard, A. Broillet, A. Chlewinski, and Th.Keller. Nachweis von Sprengstoffspuren mittelsIonenmobilitätsspektrometrie (IMS). Toxichem undKrimtech (1994) 61: 100-102

[8] A.H. Lawrence and P. Neudorfl. Detection ofethylene glycol dinitrate vapors by ion mobilityspectrometry using chloride reagent ions. Anal.Chem. (1988) 60: 104-109

[9] F. Garofolo, F. Marziali, V. Migliozzi, and A.Stama. Rapid quantitative determination of2,4,6-trinitrotoluene by ion mobility spectrometry.Rapid Communic. Mass. Spectrom. (1996) 10:1321-1326

[10] F. Garofolo, , V. Migliozzi, and B. Roio.Application of ion mobility spectrometry to theidentification of trace levels of explosives in thepresence of complex matrices. Rapid Communic.Mass. Spectrom. (1994) 8: 527-532

[11] G.E. Spangler, J.P. Carrico, and D.N. Campbell.Recent advances in ion mobility spectrometry forexplosives vapor detection. J. Testing andEvaluation (1985) 13: 234-240

[12] H.H. Hill, Jr., W.F. Siems, R.H. St. Louis, and D.G.McMinn. Ion mobility spectrometry. Anal. Chem.(1990) 62: 1201A-1209A

[13] R.H. St. Louis and H.H. Hill. Ion mobilityspectrometry in analytical chemistry. Crit. Rev.Anal. Chem. (1990) 21: 321-355

[14] Z. Karpas, Y.F. Wang, and G. Eiceman.Qualitative and quantitative responsecharacteristics of a capillary gaschromatograph/ion mobility spectrometer tohalogenated compounds. Anal. Chim. Acta. (1993)282: 19-31

[15] J. Reategui and T. Bacan. Application of ionmobility spectrometry in monitoring vapors.Process Controll and Quality (1992) 3: 209-218

[16] C. Wan, P. de B. Harrington, and D.M. Davis.Trace analysis of BTEX compounds in water withmembrane interfaced ion mobility spectrometry.Talanta (1998) 46: 1169-1179

[17] G. Eiceman. Advances in ion mobilityspectrometry: 1980-1990. Crit. Rev. Anal. Chem.(1991) 22: 471-490

[18] A.G. London, J. Kölbe-Boelke, J. Adler, and J.Stach. The use of on-site analysis for theprotection of personnel. Analusis Magazine (1995)23: 1122-1124

Th. Keller et al.: „Ion mobility spectrometry ...”, IJIMS 2(1999)1, 22-34, p. 32

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Page 37: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

[19] G. Allison, C. Saul, C.W. McLeod, and J. Gilbert.Identification of tear gases in suspect spray cansand clothes by ion mobility spectrometry. J.Forensic. Sci. (1998) 43: 845-849

[20] G. Allison and C.W. McLeod. Characterization oflachrymators by ambient temperature ion mobilityspectrometry. J. Forensic. Sci. (1997) 42: 312-315

[21] A. Miki, Th. Keller, P. Regenscheit, W. Bernhard,M. Tatsuno, M. Katagi, M. Nishikawa, L. Kim, S.Hatano, and H. Tsuchihashi. Detection of internaland external methamphetamine in human hair byion mobility spectrometry. Jpn. J. Toxicol. Environ.Health (1997) 43: 15-24

[22] A. Miki, Th. Keller, P. Regenscheit, R. Dirnhofer,M. Tatsuno, M. Katagi, M. Nishikawa, and H.Tsuchihas+i. Application of ion mobilityspectrometry to the rapid screening ofmethamphetamine incorporated in hair. J.Chromatogr. B (1997) 692: 319-328

[23] Th. Keller, A. Miki, P. Regenscheit, R. Dirnhofer,A. Schneider, and H. Tsuchihashi, Detection ofdesigner drugs in human hair by ion mobilityspectrometry (IMS). Forensic. Sci. Int. (1998) 94:55-63

[24] Th. Keller, A. Schneider, P. Regenscheit, R.Dirnhofer, Th. Rücker, and E. Tutsch-Bauer.Analysis of psilocybin and psilocin in Psilocybesubcubensis GUZMÁN by ion mobilityspectrometry and gaschromatography-massspectrometry. Forensic Sci. Int. (1999) 2: 93-105

[25] R.A. de Zeeuw, A. Kode, L. Kim, L. Cacciacarro,Hair analysis by ion mobility spectrometry, Proc.of the 1995 International Conference andWorkshop For Hair Analysis in ForensicToxicology, Abu Dhabi, (1995) 334-350

[26] L. Elias, A.H. Lawrence, On-site sampling anddetection of drug particulates, Analysis of Drugs ofAbuse, Chapter 10, T.A. Gough, Ed., (1991)373-400

[27] A.A. Nanji, A.H. Lawrence, N.Z. Mikhael, Use ofskin sampling and ion mobility spectrometry as apreliminary screening method for drug detection inan emergency room, Clin. Toxicol., (1987) 25:501-515

[28] A.H. Lawrence, Ion mobility spectrometry/massspectrometry of some prescription and illicit drugs,Anal. Chem., (1986) 58: 1269-1272

[29] G.A. Eiceman, Z. Karpas, Ion MobilitySpectrometry, CRC Press Boca Raton, Florida,1994

[30] Z. Karpas, Forensic Applications of Ion MobilitySpectrometry, Forensic Science Review, (1989) 1:104-119

[31] W.F. Siems, C. Wu, E.E. Tarver, H.H. Hill, Jr.,P.R. Larsen, D.G. McMinn, Measuring theResolving Power of Ion Mobility Spectrometers,Anal. Chem., (1994) 66: 4195-4201

[32] Valentin, Lehrbuch der Physiologie des Menschen,Braunschweig, 1844

[33] Schottin, Über die chemischen Bestandteile desSchweißes. Wunderlichs Archiv für physikal. Heilk.XI: 1852

[34] H. Tachau. Über den Übergang von Arzneimittelnin den Schweiß. Arch. Exptl. Pathol. Pharmakol.(1911) 66: 334-346

[35] H. Jansch. Zur Kenntnis gerichtlich-chemischerUntersuchungen. Beitr. Gerichtl. Med. (1922) 4:55-88

[36] G. Skopp, L. Pötsch, G. Zimmer, and R. Mattern.Zur Interpretation von Drogenbefunden auf derHaut. Blutalkohol (1997) 34: 427-434

[37] R. Fogerson, D. Schoendorfer, J. Fay, and V.Spiehler. Qualitative detection of opiates in sweatby EIA and GC/MS. J. Anal. Toxicol. (1997) 21:451-458

[38] V. Spiehler, J. Fay, R. Fogerson, D. Schoendorfer,and R.S. Niedbala. Enzyme immunoassayvalidation for qualitative detection of cocaine insweat. Clin. Chem. (1996) 42: 34-38

[39] J. Fay, R. Fogerson, D. Schoendorfer, R.S.Niedbala, and V. Spiehler. Detection ofmethamphetamine in sweat by EIA and GC/MS. J.Anal. Toxicol. (1996) 20: 398-403

[40] M. Burns and R.C. Baselt. Monitoring drug usewith a sweat patch: an experiment with cocaine. J.Anal. Toxicol. (1995) 19: 41-48

[41] E. Schneider and S. Balabanova. Nachweis vonDrogen in körpernahen Wäschestücken. Arch.Kriminol. (1991) 188: 97-105

[42] S. Balabanova and E. Schneider. Nachweis vonDrogen im Schweiß. Beitr. Gerichtl. Med. (1990)48: 45-49

[43] I. Ishiyama, To. Nagai, Ta. Nagai, E. Komuro, T.Momose, and N. Akimori. The significance of druganalysis of sweat in respect to rapid screening fordrug abuse. Z. Rechtsmed. (1979) 82: 251-256

[44] P. Kintz. Excretion of MBDB and BDB in urine,saliva, and sweat following single oraladministration. J. Anal. Toxicol. (1997) 21:570-575

[45] P. Kintz. Drug testing in addicts: a comparisonbetween urine, sweat and hair. Ther. Drug Monit.(1996) 18: 450-455

[46] P. Kintz, A. Tracqui, and P. Mangin. Sweat testingin opioid users with a sweat patch. J. Anal.Toxicol. (1996) 20: 393-397

[47] P. Kintz, A. Tracqui, C. Janey, and P. Mangin.Detection of codeine and phenobarbital in sweatcollected with a sweat patch. J. Anal. Toxicol.(1996) 20: 197-201

[48] P. Kintz, A. Tracqui, and P. Mangin. Sweat testingfor benzodiazepines. J. Forensic Sci. (1996) 41:851-854

[49] E.J. Cone, M.J. Hillsgrove, A. Jenkins, R.M.Keenan, and W.D. Darwin. Sweat testing forheroin, cocaine, and metabolites. J. Anal. Toxicol.(1994) 18: 298-305

Th. Keller et al.: „Ion mobility spectrometry ...”, IJIMS 2(1999)1, 22-34, p. 33

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Page 38: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

[50] S. Suzuki, T. Inone, H. Hori, and S. Inayama.Analysis of methamphetamine in hair, nail, sweat,and saliva by mass fragmentography. J. Anal.Toxicol. (1989) 13: 176-178

[51] P. Kintz, V. Cirimele, and B. Ludes. Codeinetesting in sweat and saliva with the Drugwipe. Int.J. Legal Med. (1998) 111: 82-84

[52] J. Jaspers. Personal communication

Th. Keller et al.: „Ion mobility spectrometry ...”, IJIMS 2(1999)1, 22-34, p. 34

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Page 39: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

AbstractIn high-voltage systems insulated with SF6

electrical discharges as partial discharges,sparks or arcs cause SF6 decompositionleading to the formation of some toxic andcorrosive by-products. There is an urgent needfor more information on the origins andquantities of contaminants expected to arisefrom the use of SF6 filled electrical powerequipment. A low resolution ion mobilityspectrometer used as analysis instrumentdelivers a shift of the position of the main peakobtained. This shift is correlated to theconcentration of decomposition productsformed in SF6. This paper presents the resultsof investigations on the fill gas in circuitbreakers in gas insulated substations duringoperation and of the reclaimed gas after arecycling procedure. Results from theinvestigation of 36 different circuit breakers inan operating substation are presented, whichcan lead to new methods to check the fill gasquality.

IntroductionIn the last 30 years the use of SF6 insulatedswitchgears in high voltage substations hasincreased considerably. This is due to theirenhanced lifetime, reduced required area andvolume and longer maintenance cycles4-6. Veryhigh standards in the engineering andmanufacturing result in the high reliability ofthese substations. However, in long termoperation failures still arise which may lead tosevere damage and additional repair costs1,3,9.Also, problems with toxic and corrosivedecomposition products and the potentialgreenhouse gas effect arising from the

artificially produced SF6 are discussed10-17. Thebasic physical and chemical properties of SF6,its behaviour under various types of dischargesand simulations of different types of equipmentused in electric power industry have beenbroadly investigated14-16. Due to its efficientinfrared absorption and chemical inertness,SF6 contributes to the stratospheric ozonedepletion. The technical and scientificcommunity and international associations suchas CIGRÉ (International Conference on LargeHigh Voltage Electric Systems) have discussedthe need for regulations relating to SF6

handling, the use of new mixtures, as well asrecycling or re-use procedures. There is nodoubt, that more information is needed on theorigins and the amounts of the contaminants inthe SF6 used in electrical power equipment.Purity standards for fresh and reclaimed SF6,as well as methods to check the gas quality arealso required. Hence, the establishment of adiagnostic tool, which allows the on-site andon-line analysis of the insulating gas containedin the gas insulated switchgear (GIS) is ofconsiderable interest. Ion mobility spectrometry is a techniquedesigned for sensitive environmentalmonitoring using low cost, rugged and highlymobile instruments. Using this technique tosupervise the insulating gas in GIS duringoperation offers the opportunity of event-basedmaintenance2,18,20-22.

Ion Mobility SpectrometryIon mobility spectrometry is a technique whichwas designed for the detection of tracecompounds within a gas, e.g. gaseouspollutants in air7,8,19,23-24. It combines high

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Monitoring of Circuit Breakers using Ion Mobility Spectrometry todetect SF6-Decomposition

J.I. Baumbach1, P. Pilzecker2, E. Trindade3

1 Institut für Spektrochemie und Angewandte Spektroskopie, Bunsen-Kirchhoff-Str.11, D-44139 Dortmund, Germany 2 G.A.S. Gesellschaft für Analytische Sensorsysteme mbH, Emil-Figge-Str. 76-80, D-44227 Dortmund, Germany3 Laboratório Central de Pesquisa e Desenvolvimento COPEL/UFPR, C.P.318, CEP 80001-970, Curitiba, PR, Brazil

Dedicated to Prof. Dr. D. Klockow on the occasion of his 65th Birthday

Received for review November 10, 1999, Accepted December 9, 1999

Page 40: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

sensitivity and relatively low technicalexpenditure with high speed data acquisition.The main advantage over other commondetection methods is the fact, that theinstrument can work on-line, continuously andunsupervised. All these advantages also applyto the application introduced here. However,there are some essential differences becauseof the application in the industry here to thetraditional operation of the ion mobilityspectrometer (IMS). This shall be made clear inthe following.The instrument is based on the drift of ions atambient pressure under the influence of anelectric field. The ions undergo a separationprocess based on various drift velocities due todifferent masses and charges on the waytowards the Faraday plate. A so called drift gas(clean air or nitrogen), which flows from the

Faraday plate to the ionization region is usedto protect the drift region from chemicalreactions between the analyte molecules andto allow only ions of the analyte formed toenter the drift region but no unchargedmolecules of the analyte. One of thepreconditions of IMS operation is that(hopefully) no charge transfer reactions occurin the drift region of the IMS. In our case, however, an application inindustrial area is considered. Therefore, nodrift gas is available in the substationinvestigated. In contrast to the commonoperation of the IMS a low resolution IMS notprotected against analyte molecules enteringthe drift region was build. Here, the real startposition of the ion swarm is unknown.Normally, a Bradbury-Nielsen shutter opens forshort time intervals between some µs and

J.I. Baumbach et al.: „Monitoring of Circuit Breakers using ...”, IJIMS 2(1999)1, 35-39, p. 36

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 1:Point to connect the circuit breaker to the ion mobility spectrometer

Page 41: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

about 1 ms and only during this time intervalsions can enter the drift region. Thus, in ourcase a flow of contaminated SF6 will enterionization and drift region. Therefore, chargetransfer reactions also occur in the drift region.The mechanism of ion formation in SF6 iscurrently under investigation using a highresolution IMS. Collecting the ions on a Faraday plate deliversa time dependent signal corresponding to themobility of the arriving ions. Normally,such an ion mobility spectrumcontains information about the natureof the different trace compoundspresent in the sampled gas. In ourcase a broad peak occurs. But theshift of the peak position could becorrelated to the total number ofdecomposition products25.

Results and DiscussionFor investigations on SF6, a table-topion mobility spectrometer (IMS) wasdeveloped and employed in aconventional substation equippedwith SF6 filled circuit breakers undernormal operation. The IMS wasconnected to the gas compartmentwith a thin polyethylene tube, whichallowed a small flux of SF6 to passthrough the instrument, adjustablewith a needle valve. To ensure

comparable conditionsreference spectra of pure SF6

were taken on-site and theyshowed no significantdifference to the spectra takenin the laboratory. It is notablethat the data acquisition time(including data processing) wasonly a few minutes under realconditions which makes thismethod especially interesting tothe potential users. The fluxwas adjusted to about 2 L/h atambient pressure. The totalmeasurement time was about 5min. The filling valve wasconnected to the IMS (seeFigure 1) with a DILO® adapterand a Teflon® line with an innerdiameter of 1 mm. A needlevalve allowed the adjustment ofa constant small flux through

the IMS. All 36 circuit breakers on the site wereinvestigated using this procedure. In Figure 2the change of the peak positions calculatedfrom the SF6 spectra of the 3 compartments ofthe phases A, B and C (former nomenclatureR, S and T) of a circuit breaker are clearlyvisible. This leads to the conclusion thatdespite the molecular sieves installed in thecompartments, significant amounts ofby-products are present in the gas of this

J.I. Baumbach et al.: „Monitoring of Circuit Breakers using ...”, IJIMS 2(1999)1, 35-39, p. 37

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 2:

Spectra obtained on the three phases of a circuit breaker anda reference spectrum of new SF6

10 20 30 40 50

0,0

0,5

1,0

Peak Position / ms Phase C 29.5 Phase B 29.1 Phase A 28.4 Reference 27.1

Peak Position

Peakshift Phase C

Mean Peak Position(5 Measurements)

Phase A (28.4+0.2) msPhase B (29.1+0.3) msPhase C (29.5+0.2) ms

Sig

nal /

V

Drift Time / ms

Figure 3:

Peak shift between the reference spectrum and thespectrum obtained from 36 different circuit breakers

A

B

C

0.0

0.5

1.0

1.5

2.0

2.5

Circuit B

reake

r

Phase

Pea

k S

hift

/ ms

Page 42: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

specific circuit breaker. The identity and originof the by-products could be investigated usingmore selective analytical methods. Theperformance of the monitoring system,however, is demonstrated through detection ofthe peak shifts (see Figure 3). The shift in thepeak position is correlated to the meanconcentration of the decomposition products.Therefore, the number of circuit breakers indifferent classes of peak shift rangescorresponds with classes of concentration

ranges (see Figure 4). However,since the distance etween the errorsource and the gas sampling point isnot known, the local concentrationscould be much higher. The CIGREvalues for the concentration (totalnumber of decomposition products)acceptable are between 500 ppmv

and 2000 ppmv. As can be seen inFigure 4 some circuit breakers wereover these limits and were thismethod to be used for eventorientated maintenance then thesecircuit breakers should be subject tomaintenance.The use of IMS, as presented here,allows an automated monitoringsystem to be set up. The structure ofsuch a system is shown in Figure 5.Crucial compartments of thesubstation should be equipped withan IMS connected to the gas insidethe compartments. The acquired

spectra could be transferred via a local areanetwork or other telecommunications system tothe control room where the IMS server isplaced. Here the spectra can be evaluated bycomparison to reference data, providing abasis for decisions on maintenance.

ConclusionsA sensitive method was developed for on-sitemonitoring of SF6 quality which proved to besuitable for the detection of small changes in

composition of the gas.Experiments carried outon-site at a gas insulatedsubstation under realconditions revealed theability of the instrumentto withstand theenvironmental conditionsoutside the laboratory.The encouraging resultsform the basis for arugged continuouslyoperating monitoringsystem for circuitbreakers in gas insulatedsubstations.

AcknowledgmentsThe financial support ofthe Bundesministerium

J.I. Baumbach et al.: „Monitoring of Circuit Breakers using ...”, IJIMS 2(1999)1, 35-39, p. 38

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 4:

Classification of the peak shift and interpretation interms of mean decomposition products expected

0.5 1.0 1.5 2.0 2.50

2

4

6

8

10

12

14

16

> 750 ppm

v

> 500 ppm

v

N

umbe

r

Peak Shift / ms

Figure 5:

Concept of a continuously monitoring system to control on-line andon-site the gas quality in different compartments of a gas insulatedsubstation

IMS

IMS

IMS

IMS

IMS

IMS

IMS-server

Power plant

Control room

Gas insulated

Monitored

compart-

of the

substation

Gas samplingand feedback

Data transfer to control room

Available network connection for system extension

ments

Connection employing existing network connections

substation

Operatingcompany

Page 43: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

für Bildung, Wissenschaft, Forschung undTechnologie and the Ministerium fürWissenschaft und Forschung des LandesNordrhein-Westfalen are gratefullyacknowledged.

References[1] Babusci, G., Colombo, E., and Speciali, R.,

“Assessment of the Behaviour of Gas-InsulatedElectrical Components in the Presence of an InternalArc”, CIGRE 1998, 21, 1998.

[2] Baumbach, J. I., Irmer, A. v., Klockow, D., AlbertiSegundo, S. M., Sielemann, S., Soppart, O., andTrindade, E., “Characterisation of SF6 DecompositionProducts caused by Discharges in Switchgears usingIon Mobility Spectrometry”, Proceedings of the fourthinternational workshop on ion mobility spectrometry,Cambridge, August 1995, 1995.

[3] Chan, T. M., Heil, F., and Kopejtkova, D., “Report onthe Second International Survey on High VoltageGas Insulated Substations (GIS) ServiceExperience”, CIGRE 1998, 23-102, 1998.

[4] Christophorou, L. G., Olthoff, J. K., and Green, D. S.,“Gases for Electrical Insulation and Arc Interruption:Possible Present and Future Alternatives to PureSF6”, NIST Technical Note, 1425, 1-44, 1997.

[5] Mauthe, G. et al., “SF6 and the global atmosphere”,Electra,164, 121-131, 1996.

[6] Moritz, G., “25 Jahre Betriebserfahrungen mitSF6-isolierten Schaltanlagen”, Elektrizitätswirtschaft,92, 1191-1194, 1993.

[7] Eiceman, G.E., Karpas, Z., “Ion mobilityspectrometry”, CRC Press, Boca Raton, Ann Arbor,London, Tokyo,1-228, 1994.

[8] Hill, H. H., Siems, W. F., St.Louis, R. H., andMcMinn, D. G., “Ion mobility spectrometry”,Anal.Chem., 62, 1201A-1209A, 1990.

[9] Neumann, C., Balzer, G., and Hudasch, M.,“Insulation Coordination Practice of German Utilities- Procedures, Experience and Future Trends”,CIGRE 1998, 33-102, 1998.

[10] Ko, M.K.W., Sze, N.D., Wang, W.-C., AtmosphericSulphur Hexafluoride: Sources, Sinks andGeenhouse Warming. - Journal of GeophysicalResearch 98 (1993) 10-499 - 10.507

[11] Maiss, M., Levin, I., Global Increase of SF6 Observedin the Atmosphere. - Geophysical Research Letters21 (1994) 569-572

[12] Van Burnt, R.J., Production rates for oxyfluoridesSOF2, SO2F2, and SOF4 in SF6 corona discharges. -Journal of Research of the National Bureau ofStandards 90 (1985) 229-253

[13] Van Brunt, R.J., Olthoff, J.K., Sauers, I., Morrison,H.D., Robins, J.R., Chu, F.Y., Detection of S2F10

produced by electrical discharges in SF6 - Proc. 10thInt. Conf. on GAS DISCHARGES AND THEIRAPPLICATIONS, Swansea 13-18 September 1992,p. 418-421

[14] Vigreux, J.: Application of condition monitoringtechniques in gas insulated substations. - Electra(1986) 48-59

[15] Mauthe, G., Pettersson, K., Gleeson, D., König, D.,Lewis, J., Molony, ; P.O'Connell; A.Porter;L.Niemeyer: Handling of SF6 and its decompositionproducts in gas insulated switchgear (GIS). - Electra136 (1991) 69-89

[16] Mauthe, G., et al., Handling of SF6 and itsdecomposition products in gas insulated switchgear(GIS) -2nd Part- Electra 137 (1991) 81-108

[17] Mauthe, G., Pyror, B.M, Niemeyer, L., Probst, R.,Poblotzki, J., SF6 Recycling Guide - Re-use of SF6

gas in electrical power equipment and final disposalCIGRE 23.10 Task Force01 23.10 (1997) 1-47

[18] Soppart, O., et al., “Ion mobility spectrometry for thecharacterization of SF6-decomposition productscaused by partial discharges”, 9th InternationalSymposium on High Voltage Engineering, Graz,August 28 - September 1, 1995, 2269-2261 -2269-2264, 1995.

[19] Roehl, R. E., “Environmental and processapplications for ion mobility spectrometry”, Appl.Spectrosc. Rev., 26, 1-57, 1991.

[20] Segundo, S. M. A., Soppart, O., Janissek, P. R.,Baumbach, J. I., Neves, E. F. A., Klockow, D., andGutz, I. G. R., “Monitoring of Sulphur Hexafluoride(SF6) - Environmental and Technological Relevance”,Proceedings of the International Colloquium onProcess related Analytical Cemistry inEnvirionmental Investgations, Gramado, Brazil, 63,1996.

[21] Soppart, O., Baumbach, J. I., Alberti, S. M., andKlockow, D., “Partial Discharge Ion MobilitySpectrometry for Rapid Quality Assessment of SF6

used in High Voltage Substations”, Field ScreeningEurope, 1997. In: J. Gottlieb, H. Hötzl, K. Huck andR. Niessner: Field Screening Europe, Proceedings ofthe First International Conference on Strategies andTechniques for the Investigation and Monitoring ofContaminated Sites, Sept. 29 - Oct. 1, 1997,Karlsruhe, Germany, p. 355-358

[22] Soppart, O., Baumbach, J. I., Alberti, S. M., Moraese Silva, J. M., and Klockow, D., “Monitoring theQuality of Sulphur Hexafluoride (SF6) in GasInsulated Switchgear sing Ion MobilitySpectrometry”, Proceedings of the V SEMEL, August1996, Curitiba, PR, Brazil, 2, 562-571, 1996.

[23] St.Louis, R. H., and Hill, H. H., “Ion mobilityspectrometry in analytical chemistry”, Critical Reviewin Analytical Chemistry, 21, 321-355, 1990.

[24] Widmer, H. M., and Morrissey, M. A.,“Neochromatographic technologies - ion mobilityspectrometry”, Chimia, 43, 268-277, 1989.

[25] Baumbach, J.I., Klockow, D., Kurrat, M., Segundo,S.M.A., Soppart, O., Verfahren zur Überwachung desQualitätszustandes des FüllgasesSchwefelhexafluorid in gasgefüllten Anlagen.German Patent DE 195 28 290. - Priority 2.8.1995

J.I. Baumbach et al.: „Monitoring of Circuit Breakers using ...”, IJIMS 2(1999)1, 35-39, p. 39

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Page 44: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

AbstractPositive ion mobility spectra of chlorobenzene,bromobenzene and iodobenzene weremeasured using different ionization processes.Corona discharge ionization andphotoionization permit a more sensitivedetection of these compounds in comparisonwith 63Ni ionization. For chlorobenzene andbromobenzene, a similar drift behavior offormed product ions can be observed. Coronadischarge ionization and photoionizationprovide one product ion peak. However, thereduced mobility values (K0 values) varydepending on the used ionization method. For63Ni ionization, two product ion peaks aredetected for these compounds. The reducedmobility values of these two peaks correspondto K0 values of the peaks observed by coronadischarge ionization and photoionization. Foriodobenzene, two peaks can be obtained forall ionization processes. However, the mostintensive peak is detected at the same reducedmobility values for all ionization processes.

IntroductionThe ion mobility spectrometry permits the traceanalysis of compounds in gaseous phase atatmospheric pressure. The most commonlyused spectrometers are equipped with 63Niionization sources [1]. Photoionization andCorona discharge ionization were developedas alternative non-radioactive ionizationprocesses.

Using 63Ni ionization, the formation of positiveproduct ions is mainly initiated by protontransfer reactions. [MH]+ ions, clustered ionsand dimers are formed depending on thetemperature in ionization region, thecomposition and moisture of carrier gas anddrift gas as well as on the chemical andphysical properties of investigated compoundsand their concentrations [2-5]. For

photoionization, the formation of [M]+ productions can be expected using UV lamps, whichsupply energies between 9 and 12 eV [6].Using Corona discharge ionization, productions can be formed through differentprocesses. The formation of positive productions may be initiated by electron impact,photoionization and proton transfer reactionsdepending on the electric field strength aroundthe corona needle [7]. Subsequention-molecule reactions or charge-transferreactions can be observed for the mentionedionization processes. Therefore, the formationof different product ions can be expecteddepending on the applied ionization process.The investigations of mono-halogenatedbenzene compounds chlorobenzene,bromobenzene and iodobenzene wereundertaken to compare the capabilities of theabove mentioned ionization sources regardingthe sensitivity of detection as well as to studythe differences in formation of positive productions.

Available experimental data for thesesubstances are based on investigations byspectrometers equipped with 63Ni ionizationsources. For mono-halogenated benzenes, theformation of [MH]+ product ions and [M2H]+

dimers is assumed in positive mode [8, 9].Similar ionization pathways are supposed forrelated compounds like dihalogenatedbenzenes and halogenated nitrobenzenes[10,11]. In negative polarity, halogencontaining ions can be mainly observed forhalogenated benzenes due to dissociativeelectron capture processes [8,12].

ExperimentalThe applied sample introduction system isdepicted in fig. 1. About 300 µl of liquidsubstances were sealed in permeation tubesconsisting of polyethylene with an internal

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Ion mobility measurements of mono-halogenated benzenes usingdifferent ionization processes

H. Borsdorf1, H. Schelhorn1, M. Rudolph1, J. Flachowsky1, J. Stach2

1 Center for Environmental Research Leipzig-Halle; Permoserstraße 15; 04318 Leipzig; Germany2 Bruker Saxonia Analytik GmbH; Permoserstraße 15; 04318 Leipzig; Germany

Received for review November 26, 1999, Accepted December 12, 1999

Page 45: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

volume of 1 ml and a wall thickness of 0,5 mm.The compounds used in this study had a purityof 99 % and were obtained from MERCK. Thepermeation tubes were placed in a permeationvessel, which was coupled with an automaticheat regulator. Purified (charcoal) and dried(silica gel with moisture indicator) ambient airwas pumped through the permeation vesselwith a constant flow rate. The sample gasstream was split by means of flow controllers.A defined volume of sample gas stream wasrarefied with purified and dried ambient air in agas-mixing chamber. A second rarefaction canbe additionally realized. A fixed constant flowrate of sample gas stream and purified anddried ambient air were primed by the samplinggas pump of IMS. The flow rate of this total gasstream into the IMS was kept constant (25 l/h).The concentration of the compounds in thesample gas stream was calculated using theweight loss of the permeation tubes overpermeation time (microbalance: Mettler ToledoMT 5).

The measurements were performed with ionmobility spectrometers manufactured byBRUKER SAXONIA. The spectrometers withbidirectional gas flow are equipped with a

membrane inlet. The temperature of inletsystem was kept at 80 °C. A krypton lamp(10,0 eV) was used for photoionizationexperiments. A 555 MBq 63Ni ionization sourcewas applied for the emission of beta particles.The details of corona discharge ionizationsource were described recently [7]. Withexception of ionization, all other operationalparameters for ion mobility measurementswere identical. The parameters used to obtainthe spectra were: carrier gas flow rate: 25 l/h;drift gas flow rate: 25 l/h; electric field: about250 V/cm; temperature of drift tube: 50 °C;pressure: atmospheric pressure. Air was usedas carrier gas and drift gas.

Results and DiscussionThe ion mobility spectra of chlorobenzene andbromobenzene are depicted in fig. 2 and 3. Acomparable drift behavior is observed for bothcompounds. Two product ion peaks areobtained using 63Ni ionization. Coronadischarge (CD) ionization and photoionization(PI) provide only one product ion peak.However, considerable differences in reducedmobility values can be observed between thepeaks obtained with these two ionizationmethods.

H. Borsdorf et al.: „Ion mobility measurements ...”, IJIMS 2(1999)1, 40-44, p. 41

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 1: Sample introduction system

pump

1st d

ilutio

n st

age

flow meters

pressurecompen-sation gas mixing chamber

2nd

dilu

tion

stag

e an

dpr

essu

re c

ompe

nsat

ion

waste

waste

gas split

perm

eatio

n ve

ssel

silicagel

pressure regulator

compensation

silicagel

pump

thermostat

pressure

charcoal

silic

agel

with

moi

stur

e in

dica

tor

Page 46: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

The reduced mobility values for the peaks withhigher K0 value obtained by 63Ni ionizationcorrespond to those observed by Karesek etal. using the sameionization method (chlorobenzene:1,99 cm2/Vs;bromobenzene: 1,91cm2/Vs; iodobenzene: 1,81cm2/Vs) and were assignedto [MH]+ product ions [6].This peak with higherreduced mobility exhibits astronger dependence onconcentration incomparison with thesecond peak detected inour investigations at lowerK0 values. Fig. 4 illustratesthe intensities of production currents in dependenceon concentration for bothproduct ion peaks ofchlorobenzene. Theintensity ratio between bothpeaks indicates, that thepeak with lower K0 value

are not caused bymonomer-dimerreactions. However,the reduced mobilityvalues of these peaksat lower reducedmobilities correspondswith those obtainedby CD ionization.Obviously, thesepeaks can beassigned to clusteredions due to ionmolecule reactions.Using PI, the K0 valueof detected singlepeaks correspondswith peaks at higherreduced mobility valueobserved in the 63Nispectra.

For iodobenzene, adeviating behavior isobserved. The peakat 1,81 cm2/Vsappears in all spectraindependent of

applied ionization process. This peak wasassigned to [MH]+ ions by Karasek et al. [6].

H. Borsdorf et al.: „Ion mobility measurements ...”, IJIMS 2(1999)1, 40-44, p. 42

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 3: Ion mobility spectra of bromobenzene using different ionizationprocesses

Br

222 µg/l

3,5 µg/l

15 µg/l

63Ni ionization(measuring range: 50 – 500 µg/l)

Corona discharge ionization(measuring range: 2 – 45 µg/l)

Photoionization(measuring range: 7 – 100 µg/l)

positivereactant ions

1,89

1,73

1,89

1,73

Figure 2: Ion mobility spectra of chlorobenzene using different ionizationprocesses

Cl

63Ni ionization(measuring range: 50 – 580 µg/l)

Corona discharge ionization(measuring range: 1,5 – 50 µg/l)

photoionization(measuring range: 10 – 100 µg/l)

165 µg/l

3,5 µg/l

20 µg/l

1,80

1,98

positivereactant ions

1,79

1,97

Page 47: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

The calibration curves are mainly influenced bythe intensity of this peak. Besides this peak,

additional product ions were detected by allionization methods. 63Ni ionization provide an

additional peak at 1,99cm2/Vs. However, thispeak can be observedwith increasingconcentrations only.

Product ion peaks withcomparable reducedmobility values areobtained using PI. Theproduct ion peak at1,94 cm2/Vs can bedetected atconcentrations > 15µg/l only. In addition tothe peak at 1,81cm2/Vs, a product ionpeak at 1,70 cm2/Vs isdetectable using CDionization. Both peaksdepend onconcentration.

Fig. 6 illustrates, thatthe use of CDionization as well as PIpermits a more

H. Borsdorf et al.: „Ion mobility measurements ...”, IJIMS 2(1999)1, 40-44, p. 43

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 4: Intensities of product ion currents in dependence on concentration

Calibration curve for chlorobenzene

50

100

0 100 200 300 400 500 600 700

concentration [µg/l]

inte

nsity

[pA

]

K(0)= 1,98

K(0)= 1,80

Figure 5: Ion mobility spectra of iodobenzene using different ionizationprocesses

I

63Ni ionization(measuring range: 60 – 760 µg/l)

Corona discharge ionization(measuring range: 4 – 25 µg/l)

photoionization(measuring range: 3,5 – 50 µg/l)

235 µg/l

6 µg/l

7 µg/l

positivereactant ions

1,81

1,70

1,82

1,82

1,99

Page 48: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

sensitive detection of mono-halogenatedbenzenes in comparison with 63Ni ionization.The measuring ranges confirm the detectionsensitivity of these ionization sources.The investigations have shown, that thenon-radioactive ionization sources can be usedfor sensitive determination ofmono-halogenated benzenes. In contrast tothe detection of negative ions, themeasurements of positive product ions permitthe assignment of substances. Defined spectracan be observed for the investigatedcompounds using the measuring conditionsdescribed above.

References[1] H. H. Hill, G. Simpson: Capabilities and Limitations

of Ion Mobility Spectrometry for Field ScreeningApplications, Field Anal. Chem. Techn., 1 (1997)

119.

[2] J.Sunner, G. Nicol and P. Kebarle: FactorsDeterming Relative Sensitivity of Analytes in PositiveMode Atmospheric Pressure Ionization MassSpectrometry, Anal. Chem, 60 (1988) 1300.

[3] G. A. Eiceman, E. G. Nazarov, J. E. Rodriguez andJ. F. Bergloff: Positive Reactant Ion Chemistry forAnalytical, High Temperature Ion MobilitySpectrometry [IMS]; Effects of Electric Field of the

Drift Tube and Moisture, Temperature, and Flow ofDrift Gas, Int. J. of IMS, 1 (1998) 28.

[4] Z. Karpas and Z. Berant: Effect of Drift Gas onMobility of Ions, J. Phys. Chem., 93 (1989) 3021.

[5] G. M. Bird, R. A. Keller: Vapor ConcentrationDependence of Plasmagrams, J. Chrom. Sci., 14(1976) 574.

[6] J. Stach, Analytiker Taschenbuch, 16 (1997) 119.

[7] J. Adler, G. Arnold, H.-R. Döring, V. Starrock and E.Wülfling in: Recent Developments in Ion MobilitySpectrometry, Ed.: J. I. Baumbach and J. Stach, Int.Society for Ion Mobility Spectrometry, Dortmund,Germany, 1998, ISBN 3-00-003676-8, p. 110-119.

[8] F. W. Karasek, O. S. Tatone, D. M. Kane: Study ofElectron Capture Behavior of Substituted Aromaticsby Plasma Chromatography, Anal. Chem., 45 (1973)1210.

[9] F. W. Karasek, O. S. Tatone: PlasmaChromatography of the Mono-HalogenatedBenzenes, Anal. Chem., 44 (1972) 1758.

[10] F. W. Karasek, D. M. Kane: PlasmaChromatography of Isomeric HalogenatedNitrobenzenes, Anal. Chem., 46 (1974) 780.

[11] T. W. Carr: Plasma Chromatography of IsomericDihalogenated Benzene, J. Chrom. Sci., 15 (1977)85.

[12] F. W. Karasek, H. H. Hill, S. H. Kim, S. Rokushika:Gas Chromatographic Detection Modes for thePlasma Chromatography, J. Chrom., 135 (1977)329.

H. Borsdorf et al.: „Ion mobility measurements ...”, IJIMS 2(1999)1, 40-44, p. 44

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 6: Calibration curves for chlorobenzene

Calibration curves for chlorobenzene

0

50

100

150

200

250

0 100 200 300 400 500 600

concentration [µg/l]

inte

nsity

[pA

]

beta ionizationphotoionizationCorona discharge ionization

Page 49: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

AbstractThe training of neural networks is assessedusing several terms and these lack precisionwith large libraries of ion mobility spectra sincethe terms are averaged for the total number oftraining spectra. Consequently, the measuresfail to show when a network has reached anoptimum in the extent of training and thedecision to halt network training becomesarbitrary and prone to error. A numeric andgraphic assessment of training by a neuralnetwork is possible when network performanceis treated as a Fermi-Dirac distribution for anextensive library of mobility spectra. This newmethod provides refined measures of networkperformance or status. In addition to providingclear criteria to prevent over-training (with largesavings in both time and committedcomputational resources), the graphic methodis also helpful in identifying flawed mobilityspectra in training sets. Modeling ofdistributions for network performance was alsoused to identify the regions of high informationcontent in mobility spectra. The region nearthe reactant ions was found to containinformation necessary for successful trainingwith mobility spectra. This unexpected resultmay disclose details yet unknown regardinggas phase ion chemistry in ion mobilityspectrometry.

IntroductionNeural networks are nearly routine tools inchemical measurements [1-3] and areparticularly useful for chemical identificationswith molecular spectra [4,5]. One essentialstep for network use is network training whendata are presented to the network forautomated learning. The training step is criticalto network performance and errors can bemade with contemporary networks if usersprematurely halt the training or if the networkbecomes over-trained, absent humansupervision and judgement. Some indicator of

the status of training or a scoring of networkquality is provided in one of several termsgiven as number scores. These include theroot mean square error (RMSE) of the trainingprocess, the RMSE from the testingexperiment, Gaussian profile analysis, andfinally the average adjusted value (AAV). Ineach of these, network performance issummarized by a number though littleinformation about the details of training iscontained in these numbers. Moreover, thesemeasures are often accessed only at the endof the training step and an additional testingstep [6,7]. Network quality and ultimateusefulness is reported only after a largeinvestment of computational time.

Other complications with these conventionalmeasures may arise when large amounts ofdata are presented to the network since theterms of performance are often averaged bythe number of spectra used for training. Thesenumerical values become indistinct whenaverages include large numbers of spectra.So long as the number of spectra is relativelysmall, these measures have been useful andproperly trained neural networks can be usedfor chemical identifications. Even withlow-resolution mobility spectra where chemicalinformation is encoded in simple spectralprofiles, networks were suitable for chemicalanalyses [8,9]. Presumably, this is becauseneural networks operate well with non-linearspectra [10-14]. Other methods, such asmultivariate analysis, have been successful forsmall libraries [15] and networks provide thegreatest promise for advanced data processingin IMS. However, improved measures ofnetwork performance are necessary and areneeded early in the training step.The use of large libraries was motivated by aninterest in making IMS a general monitor forvolatile organic compounds. A library wascreated for 204 chemicals belonging to ten

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Quantitative Assessment for the Training of Neural Networks withLarge Libraries of Ion Mobility Spectra

Erkin Nazarov1, G.A. Eiceman1, and Suzanne E. Bell2

1 Department of Chemistry and Biochemistry, New Mexico State University, Las Cruces, New Mexico 880032 Department of Chemistry and Biochemistry, Eastern Washington University, Cheney, WA, 99004-2431

Received for review December 10, 1999, Accepted December 17, 1999

Page 50: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

chemical classes of volatile organic chemicals[5] and 5-10 spectra at various concentrationsfor individual chemicals were obtained at pointson the elution profile from a gaschromatography-IMS experiment. Thus, thelibrary was comprised initially of nearly 2500mobility spectra such as those shown for4-methyl cyclohexanone in Figure 1. Fourdistinct peaks can be seen in the mobilityspectrum as the reactant ion peaks (RIP, twopeaks at input nodes 140-160, 4.1-4.4 msec),a protonated monomer M·H+ (input nodes180-200, ~5.6 msec), and a proton bounddimer, M2·H+ (input nodes 250-270, ~7.2msec). Other spectra were comparable inpeak shape, resolution of detector response,and drift time scales.

Such large numbers of spectra suggest thatmodels of distributions might be a reasonableapproach to characterizing networkperformance. The objective of this work wasthe creation of new diagnostic tools forassessing neural network behavior with largelibraries of mobility spectra. These tools,

created initially by fitting results to distributionformulas, can be presented in graphs that areconvenient to interpret or evaluate. Central tothe objective of this effort was an interest inobtaining measures of performance, early inthe training step, to conserve time andcomputational resources. Measures wereneeded also to permit delineation betweenerrors within network configurations or defectsin data presented to the network. In the nextsection, the conventional terms forperformance and the foundations for thegraph-based tools are described.

BackgroundNeural networks, as used here, are created inseveral steps including collecting spectra,processing spectra, dividing the library intotraining and testing files, training the neuralnetwork, testing the network, and finally,analyzing the results. Ion mobility spectra werepresented to an input layer consisting of 243individual input nodes. The values entered intothe input nodes were the digitized signalintensities and each input node was associated

E. Nazarov et al.: „Quantitative assessment for the training ...”, IJIMS 2(1999)1, 45-60, p. 46

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 1:Ion mobility spectra of 4-methyl cyclohexanone at various sample concentrations as createdduring a chromatographic elution profile. Spectra are arranged sequentially, from bottom totop, as the concentration rises and then falls during a chromatographic elution profile whichresembled a Gaussian distribution in vapor concentrations over a 15 second duration. Spectra were acquired every second and shown from bottom to top with an off-set for clarity.

0

1000

2000

3000

4000

5000

6000

50 100 150 200 250 300 350

Bin Number

Inte

nsity

(arb

itrar

y un

its)

Page 51: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

to a drift time (see Figure 1). The desiredoutput from the network, by chemical class,was coded into an eleven-digit word: one foreach of ten chemical classes and one code forspectra of only the background reactant ionpeaks, i.e. signal without chemical. Forexample, n-alkane spectra were coded as 1 00 0 0 0 0 0 0 0 0, spectra for cycloalkaneswere coded as 0 1 0 0 0 0 0 0 0 0 0, and thebackground spectra were coded as 0 0 0 0 0 0 0 0 0 0 1. During the training step,the network might assign a spectrum forn-heptane a value of 0.8 in the first digit and avalue of 0.2 in another digit place. All otherdigits will be assigned values of 0. Thus, theresult for training of this spectrum forn-heptane could be0 0.8 0 0.2 0 0 0 0 0 0 0. During the trainingstep, all spectra from the training set (Ntrain) arepresented iteratively to the input layer.Weightings between processing elements areadjusted by the software program until aminimum (or threshold level) is achieved for thedifference between the desired codes (forn-hexane this would be 1 0 0 0 0 0 0 0 0 0 0)and the actual value for the output codes. Thedifference is measured for all codessimultaneously and is calculated using everydigit in the output codes. This is expressedmathematically as the RMSE as shown inEquation 1:

The RMSEtrain was calculated continually andwhen the RMSEtrain reached a given value orthreshold (usually a value of 0.001), trainingwas stopped and the network was said to haveconverged. The number of iterations (or steps)necessary for network convergence wasinfluenced by Ntrain, number of processingelements in layers, library quality, and others.The number of iterations is reflected practicallyin the time necessary for convergence. Oncethe training step was finished and all theinternal weightings of the network wereestablished, spectra from the test set (Ntest)were presented to the network and theRMSEtest was calculated per Equation 2:

Thus, the RMSEtest provides a partial andquantitative measure of the success ofcomputations for a network. Anotherquantitative measure, similar to the RMSEtest,of network quality is the average adjustedvalue (AAV), described below; both AAV andRMSEtest are reported at the end of a trainingstep.

In calculating an AAV, only the important digit(Ov) from the 11-bit digit code is summarizedfrom test results for spectra for a givenchemical group or class. For example, theimportant digit in the test results for the alkanecode (1 0 0 0 0 0 0 0 0 0 0) is the first digit andthat for the background spectrum code (0 0 0 00 0 0 0 0 0 1) is the last digit. The Ov valuescan range from 0 to 1.00 and these values aresummed for all spectra tested and the sum isdivided by Ntest to obtain the AAV per Equation3:

(3) j=Ntest

AAV = Σ(Ov)j/Ntest

j=1

where j is the current number of test spectra inthe test file. The AAV also provides anumerical appraisal of network performancebut provides no further detail on why or whereflaws arose in the network. This aspect ofconventional terms is particularly limiting with alarge library (vida infra).

Experimental

InstrumentationIon mobility spectra were obtained from two ionmobility spectrometers including alaboratory-grade gas chromatograph/ionmobility spectrometer [16] and a hand-held gaschromatograph/ion mobility spectrometer, theEnvironmental Vapour Monitor or EVM [17].The EVM (Graseby Ionics, Ltd., Watford, UK,and Femtoscan, Inc., Salt Lake City, UT) wasoperated without modification and wasequipped by the manufacturer with a 2 meterOV-1 capillary column and an IMS drift tubecontaining an ionization source of 10 mCi of63Ni. The EVM was operated continuously withvapor samples taken automatically every 90s(a period longer than the elution time for themost strongly retained compound). The IMS

E. Nazarov et al.: „Quantitative assessment for the training ...”, IJIMS 2(1999)1, 45-60, p. 47

Copyright © 1999 by International Society for Ion Mobility Spectrometry

(1)RMSEtrain = (Σ(desired code-actual output)2/Ntrain)

½

(2)RMSEtest =(Σ(desired code-actual output)2/Ntest)

½

Page 52: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

analyzer was operated at 40Hz and signal wasprocessed by digital signal averaging using aninterface card and software packagecommercially available (Graseby Ionics, Ltd.)as the Advanced Signal Processor.Temperatures were 70°C for the gaschromatographic column and near ambient (ca.40-50°C) for the IMS drift tube. The other IMSinstrument included a Hewlett-Packard model5880 GC equipped with a 30m RTX-50capillary column (Restek Corp., Bellefonte, PA)and with a high temperature IMS drift tube [16]operated at 250°C. The drift gas was air andwas extensively scrubbed over molecularsieves to a moisture level of 200 ppb. Duringan experiment, signal was acquired continuallywhile the GC oven temperature was rampedfrom 50 to 200°C at 5°C/min. The injectionport and a heated transfer line between the GCoven and the IMS drift tube were 200°C. Thedata acquisition hardware and software wasidentical to that used with the EVM.

Chemicals and SolutionsOrganic chemicals representing ten functionalgroups or classes were obtained in highestpurity from various manufacturers and stocksolutions were prepared in methylene chlorideover a range of concentrations from 1-100ng/ml. Concentrations were chosen so theresidual RIP would exist throughout ameasurement assuring conditions in the ionsource that were not saturated and providingtypical concentration-dependence as thatshown in Figure 1. All solutions were screenedby gas chromatography-mass spectrometry,before creation of the IMS library, to insurecorrect identity and purity.

ProceduresIn a GC-IMS experiment to obtain ion mobilityspectra, individual solutions were injected andspectra were automatically collected andstored as spectral files. Afterwards, spectrawere extracted from ASP-format and convertedto an Excel spreadsheet format. Thosespectra with discernable response to samplevapor were formed into a single spreadsheetby chemical class. The numbers of chemicalsby class were: 14 alkanes, 13 cycloalkanes, 15alkenes, 20 aromatics, 28 alcohols, 18aldehydes, 21 ketones, 31 esters, 28 amines,and 16 organophosphorus chemicals. In all,eleven groups including the background

spectra comprised the library from the lowtemperature IMS analyzer. This libraryconsisted initially of ~2,500 spectrarepresenting 204 compounds (straight-chained,branched, and cyclic) with molecular weightsranging from 30 to 266 g/mole. The library wasrefined to 1,396 spectra by identifying andremoving spectra with insufficient intensity orspectra which had been corrupted by residualsolvent vapors [5]. All remaining spectrashowed sufficient vapor concentrations to yieldsignal-to-noise ratios >3:1 but levels werebelow source saturation; here the highestvapor concentrations caused no more than a~60% reduction in the initial intensity of thereactant ions. The library created at lowtemperature (50-70°C) was used in all neuralnetwork studies except that last, where thelibrary for high-temperature mobility spectrawas employed. Procedures for creating andpreparing spectra for the library with the hightemperature IMS analyzer were identical tothose used for that with the EVM.

Software, Computers, and ComputationalProceduresData collection, library creation, and neuralnetwork experiments were made usingcomputers with various configurations of IntelX86 processors and speeds up to 200 MHz.Spreadsheet functions were performed usingMicrosoft Excel and Quattro Pro. Spectralanalyses and distribution plots were createdusing PeakFit 4.0 (Jandel Scientific, SanRafael, CA) and neural network programmingwas performed using NeuralWorks(NeuralWare, Pittsburgh, PA).

The number of input nodes per spectrum waslimited to 243 since some spreadsheetcolumns (total number 256) were reserved forlabeling and coding the desired output values.The library was divided into training and testingspectra and this was accomplished byselecting randomly within the library one or afew spectra from each chemical family. Thetesting spectra amounted to approximately10-15% of the original library and the networkwas not allowed to see these spectra duringthe training stage. After the network wastrained, the testing set of spectra waspresented to the network and assignments ofidentity or chemical classification were made.

E. Nazarov et al.: „Quantitative assessment for the training ...”, IJIMS 2(1999)1, 45-60, p. 48

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Page 53: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

In all experiments, the process of extracting atest file, training the network and submitting thetest file for analysis was repeated 3-5 times,each time with a unique collection of trainingand testing spectra, to minimize randomresults. Average results are presentedthroughout the findings discussed below.Back-propagation neural networks were usedexclusively with a hyperbolic tangent transferfunction based on network theory as applied tochemical problems [9-11].

After the initial studies concerning the fitting ofnetwork distributions to formulas and themeaning of graphs for diagnostic uses, threeexperiments were made in order to explore thefeatures of these tools. In the first twoexperiments, a low temperature library wasemployed to examine the influence ofcorrupted spectra in the training and testingdata sets. In addition, this experiment allowedan evaluation of the response of graphicdiagnostics to different formats for presentingdata (i.e. log versus linear scaling of intensity)to the input nodes. In a thirdexperiment, the diagnostic toolwas use to probe the location ofhighest information density inspectra from a high temperaturelibrary.

Results and discussion

Graphical Presentations andAssessing Neural NetworksThe use of a single term, eitherRMSE or AAV, to appraise thebehavior of a neural network witha large library provides few, if any,insights into neural networkcomputations. This limits attemptsto refine network parameters orthe input data. While dramaticallydifferent results for RMSE or AAVmight be produced with smallnumbers of training and testingspectra, in some instances, suchdifferences are diminished by anaveraging effect that occurs withlarge libraries. This implies thatRMSE and AAV can not disclosepoor quality spectra in a librarywhere the sum of Ntrain and Ntest islarge (i.e. Ntotal = Ntrain + Ntest >

1000). An approach based on modeling ofdistributions was necessary to gauge networkperformance. This was approached by sortingvalues of Ov in descending order as shown asa graph of Ov versus spectrum number inFigure 2A. An analysis of this curve suggeststhat the plot can be divided into two regions, S1

and S2. The area for S1 is a measure ofnetwork quality and S2 is a measure of thenumber of poorly recognized spectra. Thus, aratio (P) of area S1 and total area S2+S1 can becalculated from the graph and is given byEquation 4 where P can range from 0 to 1:

(4)P = S1/(S1 + S2)

In this convention, a perfect network and librarywould yield P = 1. The term S1, the areaunder the distribution, is equivalent to ΣOv andthe total plot area (S1 + S2) is ΣOv*Ntest; so, fora perfect network where Ov=1, the total area isrelated to Ntest per Equation 5:

(5)(S1 + S2) = ΣOv*Ntest = 1*Ntest

E. Nazarov et al.: „Quantitative assessment for the training ...”, IJIMS 2(1999)1, 45-60, p. 49

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 2:Graphic distribution of adjusted values (Ov) from trainingneural networks in actual (A) and simulated (B)experiments. The curve (A) was made using adjustedvalues arranged in descending order for an ion mobilityspectral library. The plots (B) are from simulations forconstant AAV (0.9) and differing distributions of Ov.

0,0

0,2

0,4

0,6

0,8

1,0

1,2

0 50 100 150

N (test spectra)

Ov

S2

S1

A

0

0,2

0,4

0,6

0,8

1

0,00 0,20 0,39 0,58 0,77 0,96

N/Ntest

Ov

1

2

3

∆Ν

B

Page 54: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

Consequently, P can be seen to be the sameas AAV per Equation 3. Thus, the graphicalapproach preserves information contained inthe traditional methods for evaluating neuralnetwork computations and includes newinformation regarding the distribution of theneural network performance. For example,three simulations of neural networkperformance where AAV values were identical(AAV=0.9) have apparent differences indistributions of Ov = f(N/Ntest) as shown inFigure 2B. This additional and new information

provides clues to the quality either of thenetwork configuration or of library (i.e. spectracontained in the library). The sharp drop in plot1 in Figure 2B demonstrates perfect networkperformance and good and poor qualityspectra are clearly recognized. In contrast,results in plot 3 do not permit a distinctionbetween some spectra (see ∆N) as good orpoor and suggest flaws either in the network orin the library. However, all the Ov values in plot2 are greater than 0.5 suggesting that the

network recognized all spectra as goodquality, i.e. weakness resided with thenetwork configuration.

In the sections below, the importanceand validity of such conclusions areexamined although a first inspection ofgraphic diagnostics allows someconclusions pending furtherexamination of the meaning of ∆N. Oneconclusion is that this graphicpresentation of neural network behaviorwith a large library preserves traditionalmeasurement terms and offers newinsights over RMSEtest and AAV alone ortogether.

Modeling of Distributions from NeuralNetwork TrainingThe shape of the distribution curveobtained experimentally with actual ionmobility spectra (where Ntest =161 andNtrain = 1235) as shown in Figure 2A issimilar to a Fermi-Dirac distributiondescribing the energy states offermions, e.g. low energy (e-µ < kT) freeelectrons in metals. The Fermi-Dirac

distribution (Eq. 6) describes theprobability of populations of the fermionquantum state, f(e), as a function offermion energy [18,19].

In Equation 6, τ = kT and µ = chemicalpotential. When kT ~ 0, then f(ε) = 1 forε < µ and f(ε) = 0 for ε > µ. When ε-µ>> kT, the Fermi-Dirac distributionreduces to a Maxwell-Bolzmann

E. Nazarov et al.: „Quantitative assessment for the training ...”, IJIMS 2(1999)1, 45-60, p. 50

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 3:Relationship between appearance of Ov distributionsand terms modeled after a Fermi-Dirac type formula.Plots (A) are shown for s = 0.2 with different values ofN0.5 (0.1-1 in steps of 0.1). Plots (B) are shown forN0.5 = 0.95 with different values of s (0.01, 0.02-0.12in steps of 0.02). The 1st derivatives (C) are shownfor each plot (B).

-30

-25

-20

-15

-10

-5

0

N/Ntest

0

0,2

0,4

0,6

0,8

1

1,2

N/Ntest

A

0

0,2

0,4

0,6

0,8

1

1,2

N/Ntest

B

C (6) 1

f(ε) = ------------------------ exp[(ε-µ)/τ] + 1

Page 55: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

distribution, f(e) = exp[(µ-ε)/kT]. When ε-µ <<kT, the equation describes a condition whereone of two state can be occupied by particles(fermions). Where ε = µ the distribution has avalue of 0.5, corresponding to the half-integralspins associated with fermions. While theseformulas have origins in physics, the same typeof distribution can be seen in this work from theplots of Ov versus spectrum number.

(7) 1

f(N) = ------------------------- exp[(N-N0.5)/σ] + 1

where N is the current test spectrum, N0.5 is thenumber of spectra in the plot above the pointwhere Ov = 0.5, and σ is characteristic of thewidth of the distribution. Dependencies ofFermi-Dirac distribution on N0.5 and σ areillustrated in Figures 3A and 3B. When σ isheld constant, the influence of the values forN0.5 in Equation 7 is shown in the position ofcurves in Figure 3A. This demonstrates thatthe position of a curve can be referenced tothe magnitude of N0.5. In Figure 3B, theinfluence of σ is demonstrated where smallvalues of σ display abrupt changes in profilesand large values for σ give gentle slopingcurves. Consequently, the first derivatives ofthe curves in Figure 3B show a variation in thedistribution width as shown in Figure 3C. Forall spectra with N < N0.5, the values of Ov will begreater than 0.5 as expected for spectra ofgood quality. Further, if σ is large, then thenetwork is not able to recognize some spectra,and these spectra are presumably poor quality. Thus, these graphs permit delineationbetween good and defective spectra whichthen could be removed from the training set orsubjected to further scrutiny. The term N0.5 canbe obtained conveniently from the first andsecond derivatives of Equation 7 as shown inEquations 8 and 9.

When the second derivative is set to zero, thenN = N0.5 and a precise determination can bemade for N0.5 where the curve crosses the

x-axis. Consequently, this is the preferred toolfor obtaining an accurate value of N0.5.Similarly, the nature of σ can be obtained andused to evaluate network performance. Tocalculate σ, Equation 7 is rewritten perEquation 10:

Then, solving for arbitrary values of N1 and N2

as shown in Figure 4, two formulas (one for Ov1

and one for Ov2) can be written and arrangedper Equation 11:

and the value of σ can be solved per Equation12:

Thus, the slopes in curves from Figure 3Bappear as peaks of various widths (Figure 3C)and are directly related to σ. Numerical valuesfor σ become a quantitative measure of theslope in f(N). A commonly used measure ofpeak widths is the full width at half maximum(FWHM). The relationship between s andFWHM can be derived from Equation 11. ForEquation 7 when N = N0.5, the maximum valueof f’(N) can be obtained per Equation 13:

The FWHM can be determined using the halfmaximum value of f’(N0.5). If Equation (8) isequated to this value (-1/8 s), the terms[exp(N1-N0.5)/σ]=3-√8 and [exp(N2-N0.5)/σ]=3+√8 can be calculated. Consequently theinterval between N1 and N2 (the FWHM) can beobtained per Equation 14:

E. Nazarov et al.: „Quantitative assessment for the training ...”, IJIMS 2(1999)1, 45-60, p. 51

Copyright © 1999 by International Society for Ion Mobility Spectrometry

σ = (N1-N2)/ln[(Ov2/Ov1)*(1-Ov1)/(1-Ov2)] (12)

f'(N) = (-1/σ) [exp(N-N0.5)/ σ]/ (8){exp(N-N0.5)/ σ] + 1}2

f''(N)= -(1/σ)2[exp(N-N0.5)/σ]* (9) {[exp(N-N0.5)/ σ]-1}/{[exp(N-N0.5)/ σ]-1}3

exp[(N-N0.5)/σ] = [1-f(N)]/f(N) (10)

exp[(N1-N2)/σ] = (Ov2 /Ov1)* (11) [(1-Ov1)/(1-Ov2)]

f’(N0.5) = -1/4 σ (13)

Page 56: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

Thus, Ov distributions modeled after theFermi-Dirac formula provides a graphicalmeasure of neural network performance inquantitative terms that can supplement existingevaluation tools such as AAV and RMSEtest.Moreover, parameters N0.5 and σ (or FWHM)from the distribution formulas offer quantitativeassessments of the network both in the trainingand the testing steps.

Preliminary Findings on GraphicDiagnostics based on the Fermi-Diracformula for Network Processing of MobilitySpectraThe first application of graphic diagnosticsbased on models adapted from the Fermi-Diracformula to a neural network was for monitoringthe training process with a large IMS library.The results from this experiment are shown inFigure 5 where values for N0.5, AAV, σ,RMSEtest, and FWHM are plotted againsttraining epoch (or time). The training lasted forseveral days and the results were obtained atintervals of ca. 15 X 104 training epochs. Ateach epoch interval, the network wastested and distribution plots wereconstructed; from these, quantitativevalues for the new diagnostic parameterswere recorded. This was repeated untilthe training converged according to theRMSEtrain criteria of 0.001. The top twotraces show the trend for N0.5 (solid line)and the AAV (dotted line). These resultsdemonstrate that N0.5 rapidly attained afinal value, but the AAV required a longtime to come to a stable value, and onlyasymptotically approaches N0.5. Thus,N0.5 reached a conclusion within 300,000training epochs while the AAV requiredlong times, in excess of 2,000,000epochs, to approach N0.5. Practically, anassessment of the ultimate performanceof a network and library is possible usingN0.5 in only few hours, in contrast tonearly two days for the conventional term,the AAV.

Plots are shown for the FWHM (solidline), σ (dotted line), and RMSElearn (dash

line) as the bottom three curves of Figure 5.The RMSElearn rapidly falls between 0 and400,000 epochs and thereafter undergoes onlyminor changes. In contrast, the plots for theFWHM or σ demonstrate that trainingcontinued to occur thus providing an enhancedmeasure of the direction of training. In thisinstance, effective training appeared to stop at1,500,000 epochs, after which, further trainingtime was unprofitable. These resultsdemonstrate that N0.5 and FWHM (or σ) aresuperior to existing parameters of AAV andRMSElearn. Further details may be extractedfrom the original distribution curves collected atthe selected epochs.

The distribution plots of OV versus N (fromwhich N0.5, FWHM, and σ were obtained)illustrate some subtle details on networkperformance. The results from distributions atthree epochs (see Fig. 5) are shown in Figure6 and include the untreated curve (top frame),the distributions after smoothing bySavitsky-Golay algorithm (middle frame), andresults from curve fitting to a Fermi-Diracmodel (bottom frame). The untreated andtreated curves appear very much the sameuntil after the curves are processed. The firstand second derivatives of the smoothed and

E. Nazarov et al.: „Quantitative assessment for the training ...”, IJIMS 2(1999)1, 45-60, p. 52

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 4:Simulated distribution (A) and 1st derivative of thedistribution (B) for a curve fit to an adaptedFermi-Dirac formula. The full width half maximum(FWHM) of the 1st derivative is shown.

FWHM

f(N)

N1

Ov1 Ov=0.5

Ov2

N0.5 N2

f'(N)

-1/8sigma

-1/4sigma

0

FWHM = N2-N1 = (14) σ ln[(3+√8)/[(3-√8)] ≈ 3.53σ

Page 57: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

modeled curve (Figures 7) show cleardifferences. The first derivative for thesmoothed data showed oscillations which mustbe regarded as erroneous and are presumablydue to flaws in the smoothing algothrim (sincethey are not noticeable by visual inspection ofthe smoothed data in Fig. 6B). Suchoscillations also arise in the second derivative(Figure 8A) and greatly complicate an accuratedetermination of quantitative parameters ofdistributions. Thus, smoothing is disqualifiedas a means of processing. In contrast, themodeled curves exhibit reasonable andsmooth behavior (Figures 7 and 8B) fromwhich parameters of σ and N0.5 were obtainedaccurately (and used in Figure 5).

The 1st and 2nd derivative curves fordistributions of Ov versus N at the threeepochs (15, 120, and 198 X 104 epochs) areshown in Figure 9. The 2nd derivative curve(Fig. 9B) shows with clarity the point ofintersection of the plot and the x-axis, namelyN0.5. At the first epoch, N0.5 was 0.78 and wasfar from values at the other epochs. The otherN0.5 values both were 0.94 at epochs of1,200,000 and 1,980,000. Although theestimated precision of N0.5 was ±0.01, thedifferences may provide subtle informationabout the network condition, it is likely thatafter 30 X 104 epochs (see Figure 5),differences in N0.5 were random.

Comparisons of GraphicDiagnostics versus ConventionalMeasuresThe new diagnostics operate withlarge libraries of mobility spectra yetare sensitive to the manner ofpresentation of data in the trainingstep. In Figure 10, graphs areshown from two trained neuralnetworks with a common library andtwo different methods ofpre-processing of spectra. In onemethod (top plots), the absolutespectral intensities in mV were usedwhile auto-scaled values of intensitywere used in a second method(bottom plot). In auto scaling, thespectra were normalized with respectto the intensity scale. In this,average (Xav) and standard

E. Nazarov et al.: „Quantitative assessment for the training ...”, IJIMS 2(1999)1, 45-60, p. 53

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 5:Measures of performance throughout the training of aneural network. Measures (labeled) are plotted againsttraining epochs and include N0.5, AAV, FWHM, σ, andRMSEtrain.

0

0,1

0,2

0,3

0,4

0,5

0,6

0,7

0,8

0,9

1

0 15 30 45 60 75 90 105 120 135 150 165 180 195 200

Training Epochs x 104

Par

amet

er V

alue

N0.5

AAV

FWHMRMSE

Sigma

Figure 6: Distributions of Ov at three epochs for untreateddata (A), smoothed data (B), and results from aFermi-Dirac simulation (C). In each frame, plotsare taken from epochs of 1) 150,000, 2)1,200,000, and 3) 1,989,545.

0

0,4

0,8

1,2

0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1N/Ntest

Ov

1

2

3

A

0

0,4

0,8

1,2

0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1

N/Ntest

Ov

1

2

3

0

0,4

0,8

1,2

0,01 0,21 0,41 0,61 0,81

N/Ntest

Ov 1

23

B

C

Page 58: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

deviations (sd) were calculated for each of the245 input nodes in spectra from all spectra inthe library (N=1,396). Data points in thespectra were normalized (Xnormal) in each bin(Xinput node value) according to Equation 15:

While the RMSEtest values for these twonetworks were comparable (see Table 1) andthe AAV values differed somewhat, the FWHM

values (from thin solid lines) were dramaticallydifferent. These results expressed in FWHMand s, show clearly (in contrast to RMSEtest andAAV) that a problem arose in training step.This was attributed to loss of information in theintensity axis of the spectra from apre-processing step of auto scaling. This isreasonable since a log compression of theintensity scale may be expected to lead tolosses in some spectral detail.

E. Nazarov et al.: „Quantitative assessment for the training ...”, IJIMS 2(1999)1, 45-60, p. 54

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 7:1st derivative plots of Ov distributions for the smoothed (A) and fitted(B) curves as shown in Figure 6.

-12

-10

-8

-6

-4

-2

0

2

0,00 0,20 0,40 0,60 0,80 1,00

N/Ntest

1st d

eriv

ativ

es, s

moo

thed

1

2

3

A

-12

-10

-8

-6

-4

-2

0

0,20 0,40 0,60 0,80 1,00

N/Ntest

1st d

eriv

ativ

es, F

erm

i

1

2

3

B

Xnormal = (Xinput node value - Xav)*sd-1 (15)

Page 59: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

A second example of these enhanceddiagnostics is shown in Figure 11 with twoidentically configured neural networks thatwere trained uniformly with one exception. Inthis exception, the replicate network wasaltered by removing spectra regarded as lowquality because vapor pressures were low,product ion peaks overlapped with the reactantion peak or spectra corrupted bychromatographic proximity to solvent. Ininstances where solvent and sample co-existedin the IMS ion source, the ion mobility spectrawere altered through changes in the gas phase

ion-molecule chemistry necessary for ionizationof sample. Thus, an original library and arefined library (with ~2% of all spectraremoved) were presented to two networksoperating under identical network parameters.The results in Table 2 suggested that theRMSEtrain and RMSEtest were indistinguishable,as were other common parameters, betweenthe networks. However, once again the FWHMand σ showing a difference of ca. 30%between the refined library versus the originallibrary. In this example, the network wasimproved by minor refinements in a large

E. Nazarov et al.: „Quantitative assessment for the training ...”, IJIMS 2(1999)1, 45-60, p. 55

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 8:2nd derivative plots of Ov distributions for the smoothed (A)and fitted (B) curves as shown in Figure 6

-200

-150

-100

-50

0

50

100

150

200

0,00 0,20 0,40 0,60 0,80 1,00

N/Ntest

2nd

deriv

ativ

es, s

moo

thed 1

2

3

A

-200

-150

-100

-50

0

50

100

150

200

0,20 0,40 0,60 0,80 1,00

N/Ntest

2nd

deriv

ativ

es, F

erm

i

1

2

3

B

Page 60: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

library and the results demonstrate sensitivityof the graphic method.In a final example of the merits of the graphicmethod, neural networks were used tointerrogate which region of an ion mobilityspectrum was essential to categorize spectraby functional groups. Spectra were drawn, forthis neural network experiment, from a libraryof IMS spectra generated at high temperature(250°C) and low moisture (150 ppb). Undersuch experimental conditions, the threereactant ion peaks were resolved in a spectrum(see Figure 12 vis-a-vis Figure 1).High-temperature mobility spectra can providepotentially more information than low

temperature spectra through improved peakshape or fragmentation of ions [20]. In thisexperiment, portions of all spectra weretruncated in three stages, as listed in Table 3and shown in Figure 12B, and the networkswere trained on these spectra with increasingamounts of spectral content progressivelyremoved. Results from training and testingthis high temperature library, with completespectra, yielded a value of 0.99 for N0.5. Thenneural networks were trained and tested usingfor the modified data sets (i.e. truncatedspectra). Results are shown in Figure 12A andTable 3 where the behaviors of each networkare shown in the distribution graphics. In the

E. Nazarov et al.: „Quantitative assessment for the training ...”, IJIMS 2(1999)1, 45-60, p. 56

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 9:Scale magnification of Figures 7 and 8 in the region ofN0.5 for three epochs.

-200

-150

-100

-50

0

50

100

150

200

0,83 0,86 0,89 0,93 0,96 0,99

N/Ntest

2nd

deriv

ativ

es 15

120200

-12

-10

-8

-6

-4

-2

0

0,83 0,86 0,89 0,93 0,96 0,99

N/Ntest

1st d

eriv

ativ

es

15

120

200

Page 61: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

first truncation, the portion of the spectrumcorresponding to the first and second RIP

peaks was deleted and N0.5 changed from0.99 (unmarked) to 0.98 (plot 1); σchange from 0.013 to 0.014. Thissuggested that this portion of thespectrum contained little information thatwas essential for network categorizationby chemical class. In a second truncation,a significant amount of information waslost as seen in plot 2 with a markedskewing versus the original result. Thiscorresponds to N0.5 of 0.89 and σ of0.058; the RMSEtest is 0.035. When thewhole of the reactant ion peak region was

removed (plot 3), the effect is seen visually in

E. Nazarov et al.: „Quantitative assessment for the training ...”, IJIMS 2(1999)1, 45-60, p. 57

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 10:Graphic presentation from neural network computationswith identical network parameters and the same ionmobility spectral library with untreated spectra (A) andspectra preprocessed by autoscaling (B).

-1,5

-1

-0,5

0

0,5

1

1,5

0,50 0,60 0,70 0,80 0,89 0,99

N/Ntest

Ad

just

ed v

alu

es

FWHM

-1,5

-1

-0,5

0

0,5

1

1,5

0,50 0,60 0,70 0,80 0,89 0,99

N/Ntest

Ad

just

ed v

alu

es

FWHM

Table 1. Comparison of network performance criteria forresults shown in Figure 10

0.390.35RMSEtest

0.230.07FWHM0.850.87N0.5

3.53.7Epochs x 1060.0450.037RMSEtrain

0.780.85AAV

AutoscaledMillivolts

UntreatedMillivolts

Page 62: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

the graphic diagnostic (N0.5 = 0.65, σ = 0.15),yet the RMSEtest actually improved to 0.024.As seen in Table 3, the value for N0.5 and AAVwere nearly the same but AAV was slightlylower, once again, than N0.5.; thus, AAV wasshown here to be useful in contrast withRMSEtrain which provided little value foranalysis. The terms N0.5 and σ can be seen asvaluable in exploring the region of ion mobilityspectra which is information rich. Subsequentstudies (not shown) revealed that ionfragments from the sample ions exist amongthe reactant ion peaks and these fragment ions

E. Nazarov et al.: „Quantitative assessment for the training ...”, IJIMS 2(1999)1, 45-60, p. 58

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 11:Graphic results from neural network training with an originallibrary (A) and a refined library (B). The plots includedcurves for the distribution of Ov values; the 1st derivative;and the 2nd derivative.

-1,5

-1

-0,5

0

0,5

1

1,5

0,01 0,19 0,37 0,55 0,73 0,91

N/Ntest

D

1st

2nd

A

-1,5

-1

-0,5

0

0,5

1

1,5

0,01 0,19 0,38 0,57 0,76 0,94

N/Ntest

B

Table 2. Comparison of network performancecriteria for results shown in Figure 11.

0.250.29RMSEtest

0.080.11FWHM

0.920.89N0.5

1.651.25Epochs x 106

0.020.02RMSEtrain

0.910.88AAV

ExcludedIncluded

Page 63: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

are critical to the assignment of a spectrum tothe proper chemical family. Chemicals for agiven class exhibited common fragment ionsunder the conditions of temperature andmoisture used for the drift tubes in thisexperiment. As such, ionization chemistry in

high temperature IMS was shown tohave a pattern of chemical behaviorsimilar to electron-impact ionizationmass spectrometry, a common anduniversally recognized technology.The loss of information about thesefragment ions caused losses in theability of the network to identifyspectra by chemical grouping.

Summary and conclusionsLarge collections of spectra data,when presented to a neural network,provided conventional measures ofperformance yielding a distributionwhich could be modeled using anadaptation of the Fermi-Diracformula, thus avoiding effects ofaveraging. The new and sensitivequantitative measures ofperformance emerged from thisapproach and included N0.5 and theFWHM of the 1st derivative of thedistribution. These tools aresufficiently sensitive to allowdiscernment between the quality oftraining data or errors in networkconstruction, unlike conventionalmeasures of performance. Sincethese graphic measures were rootedin models of distributions, a generalutility might be possible for otherapplications of neural networkswhere training data sets are verylarge. In the specific examplesdemonstrated here, the graphic

approach provided early diagnostics of ultimatenetwork performance in the training step,allowed assessments of input data quality forlarge librarys, and revealed the information-richregions of the input data.

E. Nazarov et al.: „Quantitative assessment for the training ...”, IJIMS 2(1999)1, 45-60, p. 59

Copyright © 1999 by International Society for Ion Mobility Spectrometry

Figure 12:Effect of deleting portions of ion mobility spectra on thetraining of neural networks. The distribution plots for Ov

values are shown (A) along with a portion of the ionmobility spectra where the truncation points are marked(B).

0,00

0,20

0,40

0,60

0,80

1,00

1,20

0,00 0,15 0,30 0,45 0,60 0,74 0,89

N/Ntest

Ov

1

2

3

4

RIP1RIP2

RIP3

#2

#3

#4

Product Ion

#1

Table 3:Results from training neural networks on whole and truncated ionmobility spectra using standard and graphic-based terms forassessment. Referenced to Figure 12.

0.150.0580.0150.013σ0.650.890.980.99N0.5

0.630.870.970.98AAV0.0240.0350.0190.011RMSEtest

0-830-650-50noneRange of inputnodes removed

4321Plot Number

Page 64: International Journal for Ion Mobility Spectrometry · 2019-03-10 · and Criminalistic Relevance In the case of investigations of mixtures of analytes using ion mobility spectrometers

AcknowledgementsThe financial support from NASA (grant no.NAGY-4558) and US Army Research office(grant no. DAAH04-95-1-0541) is gratefullyacknowledged. The efforts of Y-F. Wang increating the EVM library is also acknowledged.Helpful comments from David Young aregratefully acknowledged.

References[1] Blank, T.B.; Brown, S.D. Data Processing Using

Neural Networks, Anal. Chim. Acta 1993, 277,273-287.

[2] Zupan, J.; Gasteiger, J. Neural Networks: A NewMethod for Solving Chemical Problems or Just aPassing Phase?, Anal. Chim. Acta 1991, 248, 1-30.

[3] Smits, J.R.M.; Melssen, W.J.; Buydens, L.M.C.;Kateman, G. Using Artificial Neural Networks forSolving Chemical Problems, Part I, Chemom.. Intell.Lab. Syst. 1994, 22, 165-189.

[4] Ricard, C; Cachet, C; Cabrol-Bass, D; Forrest, T.P.Neural Network Approach to Structural FeatureRecognition From Infrared Spectra, J. Chem. Inf.Comput. Sci., 1993, 33, 202-210.

[5] S.E. Bell, E.G. Nazarov, Y-F. Wang, G.A. Eiceman,“Classification of Ion Mobility Spectra by FunctionalGroups Using Neural Networks”, Analytica ChimicaActa, 1999, 394, 121-133.

[6] Philip D. Wasseman, Neural Computing, Theory andPractice, Van Nostrand Reinhold, N.Y.

[7] R. Beale and T. Jackson, Neural Computing: AnIntroduction, Adam Hilger, Bristol, England, 1990.

[8] Eiceman, G.A.; Karpas, Z.; Ion MobilitySpectrometry, CRC Press: Boca Raton, FL, 1994.

[9] Zheng, P.; Harrington, P.B.; Davis, D. QuantitativeAnalysis of Volatile Organic Compounds Using IonMobility Spectrometry and Cascade CorrelationNeural Networks, Chemom. Intell. Lab. Syst. 1996,33, 121-132.

[10] Bell, S.; Wang, Y.F.; Walsh, M.K.; Du, Q.; Ewing,R.G.; Eiceman, G.A. Qualitative and QuantitativeEvaluation of Deconvolution for Ion MobilitySpectrometry, Anal. Chim. Acta, 1995, 303, 163-174.

[11] Boger, Z.; Karpas, Z. Application of Neural Networksfor Interpretation of Ion Mobility Spectrometry andX-Ray Fluorescence Spectra, Anal. Chim. Acta,1994, 292, 243-251.

[12] Wessel, M.D.; Jurs, P.C. Prediction of Redicted IonMobility Constants From Structural InformationUsing Multiple Linear Regression Analysis andComputational Neural Networks, Anal. Chem., 1994,66, 2480-2487.

[13] Boger, Z.; Karpas, Z. Use of Neural Networks forQuantitative Measures in Ion Mobility Spectrometry,J. Chem. Inf. Comput. Sci., 1994, 34, 576-580.

[14] Bell, S.; Mead, W.C.; Jones, R.D.; Eiceman, G.A.;Ewing, R.G. Connectionist Hyperprism Neural

Network for the Analysis of Ion Mobility Spectra: AnEmpirical Evaluation, J. Chem. Inf. Comput. Sci.,1993, 33, 609-615.

[15] Snyder, A.P.; Maswadeh, W.M.; Eiceman, G.A.;Wang, Y-F.; Bell, S.E. “Multivariate StatisticalAnalysis Characterization of Application-Based IonMobility Spectra”, Analytica Chimica Acta 1995, 316,1-14.

[16] E.G. Nazarov, G.A. Eiceman, J.E. Rodriguez, andJ.F. Bergloff, “Positive Reactant Ion Chemistry forAnalytical, High Temperature Ion MobilitySpectrometry”, 7th Intern. Conf. Ion MobilitySpectrometry, Hilton Head Is, SC, August 10-13,1998.

[17] Dworzanski, J.P.; Kim, M.G.; Synder, A.P.;Meuzelaar, H.L.C. Performance Advances in IonMobility Spectrometry Through Combination WithHigh Speed Vapor Sampling, Preconcentration andSeparation Techniques, Anal. Chim. Acta 1994, 293,219-235.

[18] C. Kittel, Elementary Statistical Physics, John Wiley& Sons, Inc. NY, 1958, pp 86-91

[19] F. Mandel, Statistical Physics, John Wiley & Sons,Inc. NY, 1971, pp 276-290.

[20] S.E. Bell, “Ion Mobility Spectrometry of SelectedOrganic Compound Classes”, PhD Dissertation, NewMexico State University, Las Cruces, NM 1991.

E. Nazarov et al.: „Quantitative assessment for the training ...”, IJIMS 2(1999)1, 45-60, p. 60

Copyright © 1999 by International Society for Ion Mobility Spectrometry