survey on batch-to-batch variation in spray paints: a collaborative study

Forensic Science International 229 (2013) 80–91

Survey on batch-to-batch variation in spray paints:A collaborative study

Cyril Muehlethaler a,1, Genevieve Massonnet a,1,*, Marie Deviterne a, Maureen Bradley b,Ana Herrero c, Itxaso Diaz de Lezana c, Sandrine Lauper d, Damien Dubois d,Jochen Geyer-Lippmann e, Sonja Ketterer f, Stephane Milet g, Magali Bertrand g,Wolfgang Langer h, Bernd Plage h, Gabriele Gorzawski h, Veronique Lamothe i,Louissa Marsh j, Raija Turunen k

a Ecole des Sciences Criminelles, Institut de Police Scientifique, Universite de Lausanne, 1015 Lausanne-Dorigny, Switzerlandb Chemistry Unit, FBI Laboratory, Quantico, VA, USAc Seccion de Quımica de la Unidad de Policıa Cientıfica de la Ertzaintza, Gobierno Vasco, 48950 Bilbao, Spaind INPS, Laboratoire de Police Scientifique de Lyon, 31 Avenue Franklin Roosevelt, 69134 Ecully cedex, Francee Landeskriminalamt, KT 43, Tempelhofer Damm 12, 12101 Berlin-Tempelhof, Germanyf Forensisch-Naturwissenschaftlicher Dienst, Kantonspolizei St.Gallen, Klosterhof 12, 9001 St-Gallen, Switzerlandg IRCGN, Fort de Rosny 1, Boulevard T. Sueur, 93111 Rosny-sous-bois, Franceh Bundeskriminalamt, 65173 Wiesbaden, Germanyi INPS, Laboratoire de Police Scientifique Marseille, 97 Bvd Camille Flammarion, 13245 Marseille, Francej LGC Forensics, Culham Science Centre, Abingdon, Middlesex, UKk National Bureau of Investigation, Jokiniemenkuja 4, 01370 Vantaa, Finland

A R T I C L E I N F O

Article history:

Received 20 June 2012

Received in revised form 4 February 2013

Accepted 22 February 2013

Available online 24 April 2013

Keywords:

Forensic

Paint

Spray

Batch

Microscopy

Microspectrophotometry

Infrared spectroscopy

Raman spectroscopy

Elemental analysis

Pyrolysis GC/MS

Chemometrics

A B S T R A C T

This study represents the most extensive analysis of batch-to-batch variations in spray paint samples to

date. The survey was performed as a collaborative project of the ENFSI (European Network of Forensic

Science Institutes) Paint and Glass Working Group (EPG) and involved 11 laboratories. Several studies

have already shown that paint samples of similar color but from different manufacturers can usually be

differentiated using an appropriate analytical sequence. The discrimination of paints from the same

manufacturer and color (batch-to-batch variations) is of great interest and these data are seldom found

in the literature. This survey concerns the analysis of batches from different color groups (white, papaya

(special shade of orange), red and black) with a wide range of analytical techniques and leads to the

following conclusions.

Colored batch samples are more likely to be differentiated since their pigment composition is more

complex (pigment mixtures, added pigments) and therefore subject to variations. These variations may

occur during the paint production but may also occur when checking the paint shade in quality control

processes. For these samples, techniques aimed at color/pigment(s) characterization (optical

microscopy, microspectrophotometry (MSP), Raman spectroscopy) provide better discrimination than

techniques aimed at the organic (binder) or inorganic composition (fourier transform infrared

spectroscopy (FTIR) or elemental analysis (SEM – scanning electron microscopy and XRF – X-ray

fluorescence)).

White samples contain mainly titanium dioxide as a pigment and the main differentiation is based on

the binder composition (C–H stretches) detected either by FTIR or Raman. The inorganic composition

(elemental analysis) also provides some discrimination.

Black samples contain mainly carbon black as a pigment and are problematic with most of the

spectroscopic techniques. In this case, pyrolysis-GC/MS represents the best technique to detect

differences. Globally, Py-GC/MS may show a high potential of discrimination on all samples but the

results are highly dependent on the specific instrumental conditions used.

Contents lists available at SciVerse ScienceDirect

Forensic Science International

jou r nal h o mep age: w ww.els evier . co m/lo c ate / fo r sc i in t

* Corresponding author at: Institut de Police Scientifique, Ecole des Sciences Criminelles, Universite de Lausanne, Batochime, 1015 Lausanne-Dorigny, Switzerland.

Tel.: +41 021 692 46 16; fax: +41 021 692 46 05.

E-mail address: [email protected] (G. Massonnet).1 These authors contributed equally to this work.

0379-0738/$ – see front matter � 2013 Elsevier Ireland Ltd. All rights reserved.

http://dx.doi.org/10.1016/j.forsciint.2013.02.041

http://crossmark.dyndns.org/dialog/?doi=10.1016/j.forsciint.2013.02.041&domain=pdf

http://crossmark.dyndns.org/dialog/?doi=10.1016/j.forsciint.2013.02.041&domain=pdf


mailto:[email protected]

http://www.sciencedirect.com/science/journal/03790738


C. Muehlethaler et al. / Forensic Science International 229 (2013) 80–91 81

Finally, the discrimination of samples when data was interpreted visually as compared to statistically

using principal component analysis (PCA) yielded very similar results. PCA increases sensitivity and

could perform better on specific samples, but one first has to ensure that all non-informative variation

(baseline deviation) is eliminated by applying correct pre-treatments. Statistical treatments can be used

on a large data set and, when combined with an expert’s opinion, will provide more objective criteria for

decision making.

� 2013 Elsevier Ireland Ltd. All rights reserved.

1. Introduction

Paint traces are regularly encountered in criminalistics cases, ascoated materials are all over our surroundings and paint traces areeasily transferred during contact between coated objects. Hence,the evidential value of any trace is frequently estimated within aBayesian framework, regarding the chance of finding a randommatch with the paint trace [1]. This is estimated with surveys ofpaint from different brands, analyzed by various forensic analyticaltechniques [2–5]. However, when two samples are left undiffer-entiated, it leads to questions relating to the discrimination ofpaints within a single brand (batch): can any two samples fromtwo distinct production batches be discriminated by standardforensic techniques? The industry acknowledges that adjustmentsare performed by adding small amounts of concentrated single-pigment solutions. These single-pigments solutions can be of thesame origin as initial pigments but can also be totally different.Further, a drift in controlled color is corrected using pigments ofthe opposite color to adjust the tint [6]. The differentiation ofpaints at a batch level has not been extensively studied and most ofthe previous studies focused on automobile samples [7]. Bell et al.[8] analyzed Raman spectra of architectural lilac paints, includingtwo sample sets of batches of different colors. They observed thatthe predominant source of variation is the ratios between peaks(rutile ratioed to other components). At conclusion, they point outthe need for extensive studies to establish the extent of batch-to-batch variation within the general population. Inkster et al. [9]used a complete sequence of examination on 14 batches ofarchitectural white paints (optical microscopy, MSP, FTIR, m-XRFand Py-GC/MS). One sample could be distinguished, though thecomparison steps were only qualitative. In an additional study intwo parts, Bell et al. [10,11] compared the discriminationpossibilities of FTIR and Raman spectra on resins and whitepaints. They highlighted large differences in the Raman spectra andshowed the possibility of discriminating based not only on thebands they contain (qualitatively) but also on their relativeintensities (characterizing within-group variation). This was notpossible with FTIR because the experimental uncertainty in themeasurements was of similar magnitude to the within-groupdifferences between the resins.

The present article tries to fill that gap with an extensive studyof batch-to-batch variations to determine if a quantitativedistinction between paints of the same brand can be achieved.This work was performed as part of a collaborative study from theENFSI Paint and Glass Working Group (EPG). Four color sets ofspray paint samples (23 samples in total) were collected directlyfrom the manufacturers and were certified to belong to distinctbatches. The samples were analyzed by the 11 laboratoriesparticipating in the project utilizing 6 different analyticaltechniques (e.g. optical microscopy, infrared spectroscopy, Ramanspectroscopy, pyrolysis gas chromatography–mass spectrometry,elemental analysis and microspectrophotometry). To checkreproducibility of the results, most of the analytical techniqueswere used at least twice in different laboratories. The results wereevaluated in terms of undiscriminated pairs of samples, by visualinspection of the data and by application of multivariate statistical

techniques to the infrared and Raman spectroscopies and MSPdata.

This survey will provide information about the techniques thatoffer the highest discrimination for batch to batch samples ofvarious colors. It also addresses the improvement one could expectby using multivariate statistical techniques particularly onspectroscopic data such as infrared, Raman or MSP spectra.

2. Materials and methods

The procedure described hereafter is meant to give an overviewof the project development and the procedures followed to checkthe reproducibility of results. A total of 6 analytical techniqueswere divided among all the participants. Due to the complexcomposition of paint, techniques aiming at the characterization ofdifferent components were chosen in order to have an overview ofboth organic and inorganic content.

2.1. Samples

A total of 23 spray paint samples corresponding to 4 differentcolors (black, white, papaya (special shade of orange) and red)were collected directly from manufacturers. 17 samples weredistributed to laboratories in 2010 with six more samples added in2011. All samples within a color group are commercially availableunder the same brand and type but represent different batches,meaning they were produced at the same plant but duringdifferent time periods. Table 1 presents the samples used.

Each paint sample was sprayed on glass slides from a distance ofabout 30 cm. Spray cans were shaken for 3 min before theirdeposition onto the slides. 11 independent sets of samples wereprepared by one laboratory and sent to all the other laboratories foranalysis.

2.2. Methods

Table 2 summarizes the instrumentation and parameters usedby each laboratory. No recommendations on the instrumentalconditions were given; each laboratory measured the samplesaccording to their own procedures. The pyrolysis GC/MS measure-ments were only completed for the 2010 sample set. The samplepreparations for each technique were as followed:

Microscopy: Two preparation techniques were used: (1) Paintsamples were rolled flat on a glass slide, cleaned with ethanol anddried (10 microns thickness) then they were either observed withno mounting media or mounted in Entellan with coverslip. (2)Samples were deposited on black gelatin lifts without coverslip forreflectance observations. Triplicate of all samples were used andcomparison done side by side in a single field of view.

Microspectrophotometry (MSP): Paint samples were measured inreflectance directly on their glass slides. No sample preparationwas needed.

Infrared spectroscopy (FTIR): Measures were made in transmit-tance on a small amount of paint scraped and flattened thendeposited on a KBr pellet.

Table 1Spray paint samples collected from manufacturers and analyzed during the survey. Samples marked with a * were added to the survey in 2011.

Sample code Color Manufacturer Product description Batch-number Production date

month/year

W1 White Motip Dupli RAL 9016; Verkehrweiss; 400 ml; Dupli-Color Acryllack 2100708414 7/2008




P1 Papaya Motip Dupli Platinum; Papaya sat in mat; 400 ml; finest paint for decoration and art 860308315 3/2008






P7* Papaya Motip Dupli Platinum; Papaya sat in mat; 400 ml; finest paint for decoration and art 540210420 2/2010

P8* Papaya Motip Dupli Platinum; Papaya sat in mat; 400 ml; finest paint for decoration and art 2220810400 8/2010

R1 Red True Colorz 3003; RubinRot glanzend; 400 ml K110 4/2010

R2 Red True Colorz 3003; RubinRot glanzend; 400 ml K210 7/2010

R3* Red True Colorz 3003; RubinRot glanzend; 400 ml K210 7/2010

R4* Red True Colorz 3003; RubinRot glanzend; 400 ml L230 8/2011

B1 Black Herpe/Kwasny Schwarz Matt; 122003; 600 ml J274 9/2009


B3 Black Herpe/Kwasny Schwarz Matt; 122003; 600 ml K033 2/2010

B4 Black Herpe/Kwasny Schwarz Matt; 122003; 600 ml K194 7/2010


B6* Black Herpe/Kwasny Schwarz Matt; 122003; 600 ml L189 7/2011

B7* Black Herpe/Kwasny Schwarz Matt; 122003; 600 ml L151 5/2011

C. Muehlethaler et al. / Forensic Science International 229 (2013) 80–9182

Raman spectroscopy: Depending on the laboratory procedure,paint samples were measured in reflectance directly on their glassslides or scraped and flattened then deposited on an Aluminumfoil.

Pyrolysis gas chromatography mass spectrometry (pyGCMS): thinpeels of the paint sample were taken using a scalpel blade and wereplaced into quartz pyrolysis tubes using quartz wool as a supportmedium to position the sample approximately 15 mm from thetop.

Elemental analysis: Scanning electron microscopy-energydispersive X-ray analysis (SEM-EDX): samples were removedusing a scalpel blade, deposited on a carbon holder with a thinlayer of adhesive and carbon coated by vacuum evaporation. For

Table 2List of the techniques and instruments used by the different laboratories during the su

Technique Laboratory

code

Instrument

Microscopy 1 Stereomicroscope

Fluorescence/Pola

Microspectrophotometry (MSP) 2 J&M Tidas MSP 4

resolution 1 nm

Fourier Transform Infrared (FTIR) Spectroscopy 3 Thermo Nicolet a

4 Bruker Optics an

5 Bruker Optics an

6 Thermo Nicolet a

Raman Spectroscopy 7 Renishaw InVia,

200–2000 cm�1

8 Horiba LabSpec, O

3500 cm�1

9 Renishaw InVia,

250–3200 cm�1

Elemental Analysis 10 Vega-3 scanning

time)

11 FEI Quanta (SEM

12 HITACHI S-3500N

13 microXRF Horiba

Pyrolysis GC/MS 14 Pyrolysis autosam

length, 250 mm d

15 Pyrolysis autosam

length, 250 mm d

16 Pyrolysis autosam

(30m length, 250

17 Frontier Labs Py-

length, 250 mm d

low vacuum conditions, samples were only deposited on aconductive adhesive tape and fixed on a SEM sample holder. X-ray micro fluorescence (mXRF): samples were scrapped anddeposited on an adhesive tape designed specifically for X-rayinstruments.

2.3. Comparison

Each laboratory was asked to measure at least 3 replicates persamples. For MSP, infrared and Raman, at least 7 replicates weremeasured in order to treat them statistically.

Laboratories provided the results with a list of non-differenti-ated samples based on visual comparison of the samples/spectra/

rvey.

: Low power microscope Leica MZ95 (10� to 60�) Bright Field/Dark Field/

rization: Leica DMLPDS and DMLM (4�, 10�, 50�)

00 series, Leica Microscope DM4000M, objective 20�, range 380–780 nm,

nd continuum microscope, resolution 4 cm�1, range 4000–600 cm�1, MCT detector

d Hyperion microscope, resolution 4 cm�1, range 4000–600 cm�1, MCT detector

d Hyperion microscope, resolution 4 cm�1, range 4000–600 cm�1, MCT detector

nd continuum microscope, resolution 4 cm�1, range 4000–600 cm�1, MCT detector

Leica DM 2500 M microscope, 633 nm (He–Ne) and 785 nm (NIR Diodes) lasers,

lympus BX 41 microscope, 633 nm (He–Ne) and 785 nm (NIR Diodes) lasers, 100–

Leica DM 2500M microscope, 633 nm (He–Ne) and 785 nm (NIR Diodes) lasers,

electron microscope (SEM), Apollo X SDD detector (EDS), (30 kV, 200 s, 30% dead

), EDAX Genesis SiLi detector (EDX),(10–30 kV, 100 s, 10–20% dead time)

(SEM), Oxford SiLi detector (EDX), (25 kV, 100 s, 40% dead time)

7000, XR tube (Rhodium), SiLi detector (EDS), (15–30 kV, 800 s, 4–9% dead time)

pler CDS 2500, gas chromatography Agilent 6890, column Zebron ZB 35 (30 m

iameter, 0.25 mm film thickness), mass spectrometer detector Agilent 5973 N

pler CDS 2500, gas chromatography Agilent 6890N, column Zebron ZB 35 (30 m

iameter, 0.25 mm film thickness), mass spectrometer detector Agilent 5973N

pler CDS 5250, gas chromatography Agilent 7890A, column Agilent 190915-433

mm diameter, 0.25 mm film thickness), mass spectrometer detector Agilent 5975 C

2020ID, gas chromatography Agilent 7890A, column Agilent 190915-433 (30 m

iameter, 0.25 mm film thickness), mass spectrometer detector Agilent 5975C

Fig. 1. Fluorescence observation of the papaya samples using blue light excitation

(top) and UV light excitation (bottom) using 20� objective. (For interpretation of

the references to color in this figure legend, the reader is referred to the web version

of the article.)


pyrograms. The discriminating power of the techniques wascalculated according to the Smalldon–Moffat formula [12].

Comparisons between laboratories were based on theirdiscrimination lists. No direct comparisons of the analytical datawere carried out as these conditions were chosen according to eachlaboratory’s internal procedure. The aim of this survey was to be asclose as possible to real casework conditions.

2.4. Chemometrics

In this section only the spectroscopic data (MSP, FTIR, Raman)were considered as their results should theoretically be lessdependent on analytical parameters than for other techniques likeelemental analysis and pyrolysis GC/MS. Optical examinations arenot suitable for chemometric treatments.

The goal of the chemometric analysis of the spectra is to evaluateits grouping and discrimination possibilities as compared to a visualinspection. Therefore, none of the information about the results sentby the participating laboratories was available and all treatmentswere blind-performed. In order to evaluate the intra-samplevariations, seven measurements were requested for each sample.

2.4.1. Data handling

Upon receipt of the spectra, the quality and measurementparameters were checked. First, the spectral range needed to behomogenized, which meant selecting the wavelength interval of thesmallest spectrum. When needed, interpolation of the results tomatch the x-values (wavelengths or cm�1) was performed. Then allspectra were stored in databases created with KnowItAll (Informat-ics System 8.2, Spectroscopy Edition, Bio-Rad Laboratories, Inc.).

2.4.2. Data pre-treatment

Parameters and pre-processing for the chemometric analysisare given in Table 3. The pretreatment and variable selectionparameters were chosen for systematic use on all spectra [13]. Itwas performed this way to ensure reproducibility for systematictreatment and handling of the data.

2.4.3. Principal component analysis (PCA)

Principal Component Analysis (PCA) was performed using theAnalyzeITTM MVP module for KnowItAll. The goal of PCA is torepresent the samples by new latent variables that are linearcombinations of the initial variables intensities. Combining theinitial variables in this way allows a facilitated and summarizedvisualization in a lower dimensional space. Often, groups of similarmeasurements can be highlighted and the separation can beinterpreted using loadings plots. The procedure for differentiatingthe samples was used as in [13]. After a single run of PCA, groups(or particular structures in the data set) were identified andisolated based on their 95% confidence boundaries betweenclasses. For each subgroup, additional runs of PCA were performediteratively until the data were randomly distributed (no moresubgroups present).

3. Results

The results are presented by the techniques applied, and furtherdivided by color sets.

Table 3Parameters used in the principal component analysis, pre-treatments and the variable

FTIR R

Pretreatment Standard normal variate (SNV) + 1st

derivative (5 pts)

B

n

Variable selection 600–1850 + 2700–3800 cm�1 3

3.1. Microscopy

Optical observations highlighted some differences for the whitesample W2, which exhibited less intense fluorescence under UVlight. Samples W1, W3 and W4 behaved similarly under differentillumination conditions.

For the papaya samples, visual differences were observedbetween the samples. Sample P4 presented much higher fluores-cence in both UV and blue light (Fig. 1). Another group with P6, P7and P8 could be separated from the others using dark fieldreflectance. The polarized transmitted light did not allowdiscrimination of the samples due to poor transmission throughthe paint. The differences found with the various illuminationswere significant and also visible to the unaided eye, due to theshade being darker.

The red samples R1–R4 could not be differentiated under thedifferent illumination conditions. They all exhibited the samefluorescence and polarization effects.

The black paints could not be distinguished. None of themexhibited fluorescence and they were essentially featureless underother illumination techniques.

3.2. Microspectrophotometry

The black and white samples were set aside and only thecolored samples were measured by MSP. The papaya samples P7and P8 had reflectance spectra that could be distinguished from theother samples. Characteristic features were the overall spectralshape as well as the absence of shoulders around 550 nm. IterativePCA using 1st derivative spectra allowed differentiation of all

s selection by interval.

aman MSP

aseline correction (quadratic) + Standard

ormal variate (SNV) + 1st derivative (5 pts)

1st derivative (5 pts)

00–2000 cm�1 380–700 nm

Fig. 2. Example microspectrophotometry spectra of the eight papaya samples (left), and their corresponding principal components analysis (right).

Fig. 3. Example microspectrophotometry spectra of the four red samples (left), and their corresponding principal components analysis (right).


samples except P6. This sample had much higher inhomogeneity(variation between the replicates) and was not differentiated fromsamples P1 to P3 (Fig. 2).

The red samples could be separated into two distinct groups,one with samples R1-R2 and another one with R3-R4 (Fig. 3). Thedistinction between their reflectance spectra was based on a lower

Fig. 4. Example infrared spectra of the four white samples (left), a

transmittance after 620 nm for samples R3 and R4. PCA givessimilar results with two well separated groups. A furtherdistinction between samples R3 and R4 could potentially be made.However the similarity between the replicates (intra-variability) iscomparable to the similarity between samples (inter-variability),so R3 and R4 were not differentiated from each other.

nd their corresponding principal components analysis (right).

Fig. 5. Example infrared spectra of the eight papaya samples (left), and their corresponding principal components analysis (right).


3.3. FTIR

All four color groups have a nitrocellulose and orthophthalicalkyd binder composition. The papaya and white samples alsocontain titanium dioxide (TiO2). The red samples have a couple ofpeaks that can be attributed to an organic pigment that was notidentified.

The white color group presents some differences for samplesW1 and W2 (Fig. 4). They both have an inversion in relativeintensities between 2920 and 2960 cm�1 compared to W3 and W4.Furthermore, sample W1 has peaks corresponding to styrene

Fig. 6. Example Raman spectra of the eight papaya samples (top left), and their corresp

analysis of samples P1, P2 and P7 (bottom left) and samples P3, P4 and P6 (bottom rig

(3000–3100 cm�1) and also higher intensities than W2 at 1490 and1450 cm�1. Statistical treatments with PCA on the spectra gaveresults similar to the visual comparison of the data.

The papaya samples were very similar to one another andspectral differences were within the range of variation observed forthe replicates. Only sample P6 has distinctive spectral variation,with much more intense peaks at 1335 and 1515 cm�1 (Fig. 5). Onelaboratory mentioned some differences for sample P3, withvariable relative intensities between 1460 and 1520 cm�1. Thesewere not observed by other laboratories as this variation was alsopresent between replicate measurements. The chemometric

onding principal components analysis (top right). Detailed principal components

ht). Sample P8 had a much larger baseline shift than any other sample.

Fig. 7. Example Raman spectra of seven replicates from sample R4 before (top) and

after (bottom) an automatic baseline correction.


analysis yielded comparable results to the visual comparison, withsample P6 differing from the other paints.

The red and black samples were not differentiated. Thedifferences between samples were of the same order of magnitudeas those between replicates. Chemometrics methods could notdistinguish between theses samples either.

3.4. Raman spectroscopy

The white paints contained titanium dioxide (TiO2) in the rutileform. The papaya samples also contained some rutile titaniumdioxide as well as pigment yellow 74. The red paints had pigmentred 170 but also presented a very broad fluorescence with the785 nm laser, leading to large baseline shift in the samples. Finally,the black samples gave poor quality spectra with peaks attributedto carbon black.

As with the infrared spectra, the white samples W1 and W2presented some differences that are distinct from samples W3 toW4. The general shape of the spectra are different with various

Fig. 8. Raman spectra of the four red samples, baseline corrected (top), and their corresponding principal components analysis using baseline correction (quadr.) + 1st

derivative + SNV (bottom left) and baseline correction (quadr.) + 2nd derivative (bottom right, slightly tilted to improve visualization).

Fig. 9. Example of the elemental analysis results by SEM/EDX, papaya samples (left) and red samples (right).


relative intensity modifications. W1 has some additional peaks at1154 and 1198 cm�1, while W2 has specific peaks at around 1525–1550 cm�1. PCA highlights the same spectral differences and thesame differentiation is obtained.

For the papaya samples, very large differences in the separationor grouping were observed between laboratories. P3, P6, P7 and P8were all individualized by some laboratories while left undiffer-entiated from other samples by remaining laboratories (Fig. 6).Although difficult to discriminate visually all together, pairwisecomparisons between all spectra shows the potential to distin-guish the different batches. PCA was very helpful with this sampleset, as the visualization capabilities of chemometric methods allowa quick understanding of the relative variations between samples.When performed iteratively, the principal component analysiscould separate all 8 different batches.

For the red samples, sample R4 could be distinguished due tothe presence of additional peaks at 680 and 750 cm�1, as well as at1455 and 1530 cm�1. All other samples in the color group could notbe differentiated. The chemometric analysis did not perform wellon this particular sample set. Sample R4, which truly possessesadditional peaks, was not discriminated by principal components

Fig. 10. Illustration of pyrolysis results from two laboratories. White samples W1 to

differentiated by the absence of two compounds.

analysis. A reason could be the very high fluorescence that cannotbe eliminated completely by pre-treatments (Fig. 7). The largebaseline shift consequently observed on the spectral intensitiesyielded a lower sensitivity for these statistical methods on thosespecific small peaks. In this special case, further tests with specificpre-treatments were performed in order to eliminate this broadbaseline and enhance the chemical information. Hard baselinecorrections using a quadratic fit followed by a second derivativesucceeded in differentiating sample R4 (Fig. 8).

Differentiation among black samples was very difficult due tothe poor Raman spectra quality (very noisy spectra and absence ofspectral features except carbon black). One lab found samples B1,B6 and B7 different from B2 to B5. Chemometrics was not able tofurther discriminate between these samples and the variationobserved between replicates was very similar to the inter samplevariation.

3.5. Elemental analysis

The most intense emission lines present in the elementalspectra of the different paint groups were consistent between

W4, all differentiated based on relative intensities (left). Red samples R1 and R2,

Fig. 11. Illustration of pyrolysis results of one laboratory for the black samples. Sample B4 gives a completely different pyrogram and could be differentiated from the other

samples.


Table 4Calculated discriminating power for each color group according to the Smalldon–Moffat formula [12]. Results in parentheses represent the statistical discrimination when

available. Ranges indicate more than one laboratory with different results.

Discriminating power [visual (statistical)]

Techniques White (n = 4) Papaya (n = 8) Red (n = 4) Black (n = 7)

Optical 0.5 0.68 0 0

MSP / 0.43 (0.92) 0.66 (0.83) /

FTIR 0.83 (0.83) 0.25–0.46 (0.25) 0 (0) 0 (0)

Raman 0.83 (0.83) 0.25–0.82 (0.96) 0.5 (0.5–0.83) 0–0.57 (0–0.57)

SEM–EDX/mXRF 0–0.5 0.64 0.5 0–0.47

Pyrolysis GC–MSa 0.83–1 0 0–1 0.4

Combined 0.83 (0.83) 0.78 (1) 0.83 (0.83) 0.57 (0.57)

a Pyrolysis GC–MS is not included in the combined discriminating power because it was not applied on the entire data set.


laboratories which identified the same elements. The smaller lineshowever varied more, depending largely on the instrumentalconditions, the detection limit and the interpretation made bylaboratories.

No differences were detected for the white samples W1 to W4by laboratories that utilized SEM. However, the m-XRF instrumentdifferentiated sample W2 from the others based on the relativeconcentration of titanium (Ti).

The results for the papaya samples are reproducible betweenthe different laboratories. Sample P6 differs in the intensity of thecalcium (Ca) emission line detected, while samples P7 and P8 havea much higher intensity for iron (Fe) (Fig. 9).

Sample R4 was discriminated from R1 to R3 due to a higheramount of titanium (Ti) (Fig. 9). The black samples wereundifferentiated by elemental analysis for all laboratories thatutilized SEM. The laboratory using m-XRF found samples B6-B7different from B1 to B5 (absence of sulfur (S) and calcium (Ca),presence of zinc (Zn)).

3.6. Pyrolysis GC/MS

Only a part of the samples (survey 2010) were analyzed bypyrolysis GC/MS. The pyrolysis GC/MS results differed significantlyfrom laboratory to laboratory. These differences included thenumber of compounds detected, their relative intensities, and theiridentification. Thus, depending on the analytical conditions, thediscrimination was different for each laboratory.

The white samples could potentially all be differentiated on thebasis of relative intensities for one lab (Fig. 10), while another onefound that samples W3 and W4 were similar. This last laboratoryfound that the differences between W3 and W4 were within theawaited variability for samples of the same source.

For the papaya samples no differentiation was possible; allsamples produced non differentiated pyrograms.

The red samples R1 and R2 were differentiated by one lab due inparticular to the absence of two compounds in R2 (Fig. 10). Anotherlaboratory found that these two samples were not distinguishable.

Black samples analyzed by pyrolysis GC/MS showed somediscrimination among the samples (Fig. 11). Sample B4 was

Table 5Final discrimination of the batch samples, combination of all techniques and calculated d

results were left out of these calculations, as they were not applied on the entire set.

White [DP = 0.83] Papaya [D

Discrimination W1 P1

W2 P2

W3, W4 P3

P4

P5

P6

P7

P8

differentiated by all laboratories, due to its completely differentpyrogram (presence of additional compounds and differences inrelative intensities). The 4 remaining samples presented somesmall differences in signal intensities which resulted in variousinterpretations by the laboratories. These differences can dependon sample amounts and other factors influencing reproducibilitythat are not discussed in further detail.

4. Discussion

4.1. Discriminating power of the techniques

The discriminating powers (DP) of the techniques werecalculated for each color group. The results presented in Table 4confirm that the differentiation by the various techniques isdependent on the color of the samples as well as on the type ofcompound detected (organic or inorganic).

Results indicated that methods based on the pigment/color characterization (e.g. optical microscopy, MSP, Raman)were better able to detect differences in colored sample sets(papaya and red). Their binder composition (FTIR, pyrolysis GC/MS) however, seemed more stable and only minor differenceswere detected. Elemental analysis also provides good discrimi-nation.

The white sample set presented some differences, mostly in thebinder composition (FTIR, Raman, Py-GC/MS).

The black sample set was very difficult to distinguish andpyrolysis GC/MS was the only method capable of providing somedifferentiation of these samples.

4.2. Batch-to-batch variation

As illustrated in Tables 4 and 5, differences between batches ofspray paint samples can be detected. Despite different laboratoriesusing different instrumental conditions, some trends can still beobserved among all the samples:

Colored samples are more likely to show differences betweenbatches. The techniques aimed at the characterization of pigments/colors are able to detect these differences.

iscriminating power. Underlined samples are those left undifferentiated. Py-GC/MS

P = 1] Red [DP = 0.83] Black [DP = 0.66]

R1, R2 B1

R3 B2, B3, B4, B5

R4 B6, B7


White samples present some differences in their bindercomposition.

Analytical techniques cannot easily differentiate among theblack samples.

These results indicate that detected variations are mainly basedon a pigment’s identity, concentration, or pigment mixtures(identity and proportion). This is especially the case for coloredpaints (red and papaya). For black and white paints, the mainpigment does not vary and no pigment mixture is detected. Thus,for these samples, batch variations are less easily detected.

These findings are explained by the manufacturing process ofpaint, especially because the color of the end product undergoesquality controls and is corrected or adjusted if necessary, by addingconcentrated single-pigments solutions. This control is generallymade by calculating the CIELAB coordinates of dry (or wet)reflectance MSP spectra [14].

The presence of added pigments represents an increasedpossibility of differentiation for the forensic scientist. Then, ifpresented with a case involving possible batch-to-batch differ-ences, it seems sound to use different strategies depending on thecolor of the samples.

Table 5 presents the final discrimination by a combinedsequence of examination on all samples.

4.3. Reproducibility between laboratories

Some differences between the laboratories using the sametechnique were observed. The most obvious example is the Ramanresults for the papaya samples which were all different (e.g. 3laboratories, 3 different classifications). It should be noted that theanalytical parameters used by each laboratory were not identicalbecause these were not imposed and left to their own routineusage. However, there is a substantial difference between thetechniques in terms of robustness. FTIR instruments haverepeatable (e.g. low variability between replicates of same sample)as well as reproducible results (e.g. low variability betweeninstruments on same sample). In contrast, Raman, elementalanalysis and Pyrolysis GC/MS seem less robust and were verysensitive to instrumental set up and conditions. Elemental analysiswas consistent between laboratories with only slight differenceson the same instrumentation. If the parameters are not suggestedor standardized (e.g. ENFSI’s best practice manual or SWGMAT’sguidelines) it seems illusory to be able to compare traces measuredin different locations. This also has implications in the creation ofcommon databases.

4.4. Statistical versus manual comparison of the spectra

The performance of chemometric methods greatly dependupon the analytical technique used. For infrared spectra, theperformances are very similar with the same distinctions beingmade between the samples visually or statistically (Table 4). Thediscriminating powers were comparable, which means that aftercorrect pretreatment of the spectra (eliminating all non-chemicalsystematic variation) the multivariate methods perform equally aswell as a human being. One main advantage of the statisticalmethods is that they save time and give an objective way ofdifferentiating the spectra.

For MSP spectra, the PCA greatly facilitates the comparison.Only very characteristic spectral features such as minima, maximaor shoulders are taken into account when visually comparing thespectra. Other features, such as slight deviations in slope orbaseline, are much more difficult to appreciate. For these reasons,multivariate methods represent a great advantage since PCA takesthe whole spectrum into account for the calculations and all ofthese minor differences are used in the differentiation process.

The comparison with chemometrics for Raman spectra was alsogood and often better than the visual data comparison (except forthe special case of the R4 sample). Due to the general shape ofRaman spectra (many correlated peaks on top of an irregular orinconsistent baseline) the visual discrimination between samplesis less evident than FTIR. Small relative intensity differences werenot easily detected visually, but could be with chemometrics. Thisenhanced sensitivity to small differences is however a drawback ifnot handled correctly. Before comparing spectra, there is a need toensure that all non-informative variation (e.g. fluorescence) iseliminated by applying appropriate pretreatments, becauseotherwise it will be modeled in the PCA. Only corrected spectra(free of undesired variation like fluorescence, baseline, randomnoise or instrumental) can exploit the full potential of multivariatemethods. This was the situation witnessed for sample R4 whichhad a large baseline shift that could not be totally eliminated withthe initial pretreatments. These observations suggest that furtherstudy of the pretreatments and of feature (variable) selectionmight yield improvements by reducing the amount of unwantedvariability.

5. Conclusion

This study represents the most extensive analysis of batch tobatch variation in spray paint samples to date. This question is ofgreat interest as the evaluation of chance matching between twodistinct batches has to be addressed when confronted withundifferentiated paint samples. The analysis of batches fromdifferent color groups with a wide range of analytical techniqueslead to interesting observations. Colored samples exhibited ahigher differentiation between batches because their compositionwas more complex (added pigments) and more subject to slightvariations. For these samples, techniques aimed at color/pigmentscharacterization (optical microscopy, MSP, Raman) performedbetter than techniques aimed at the binder composition (FTIR, Py-GC/MS) or global composition (elemental analysis).

On the other hand, techniques focused on binder composition(FTIR, Raman, Py-GC/MS) proved to be the best techniques todifferentiate the white samples. Black samples were problematicwith most of the spectroscopic techniques (light absorbance andcarbon black interference). Pyrolysis-GC/MS represented the besttechnique to detect differences in these samples.

Finally, the comparison between visual and statistical (PCA)discrimination showed similar results. The PCA has increasedsensitivity and could perform better on specific samples, but wefirst have to ensure that all non-informative variation (baselinedeviation) is eliminated by applying correct pre-treatments. Thesesteps need more studies in order to achieve their full potential onspectroscopic data.

Acknowledgements

Special thanks to IRCGN (France) and BKA (Germany) forproviding the samples for the study. We are also very thankful tothe manufacturers for letting us analyze their samples and shareinformation on the stability of their formulations.

References

[1] C. Aitken, F. Taroni, Statistics and the Evaluation of Evidence for ForensicScientists, 2nd ed., John Wiley & Sons Ltd., 2004.

[2] R. Gosse, S. Milet, B. Espanet, Discrimination of black spray paints, in: Proceedingsof the European Paint and Glass Group Meeting, Berlin, 2005.

[3] F. Govaert, M. Bernard, Discriminating red spray paints by optical microscopy,Fourier transform infrared spectroscopy and X-ray fluorescence, Forensic Sci. Int.140 (2004) 61–70.

http://refhub.elsevier.com/S0379-0738(13)00127-8/sbref0005








[4] F. Govaert, G. Roy, B. Decruyenaere, D. Ziernicki, Analysis of black spray paints byfourier transform infrared spectrometry, X-ray fluorescence and visible micros-copy, Probl. Forensic Sci. XLVII (2001) 333–339.

[5] P. Buzzini, G. Massonnet, A market study of green spray paints by fourier transforminfrared (ftir) and raman spectroscopy, Sci. Justice 44 (3) (2004) 123–131.

[6] W. Lemons, Color control in electrocoat, Coatings World, (October) (2006).[7] W. Stoecklein, C. Palenik, Forensic analysis of automotive paints: evidential value

and the batch problem, in: Proceedings of the 4th European Paint Group Meeting,Paris, 1998.

[8] S.E. Bell, L.A. Fido, S.J. Speers, W.J. Armstrong, Rapid forensic analysis andidentification of ‘‘lilac’’rchitectural finishes using raman spectroscopy, Appl.Spectrosc. 59 (1) (2005) 100–108.

[9] J. Inkster, P. Maynard, C. Roux, B. Fergusson, Intrasample vs intersample variabili-ty in architectural paint, in: Proceedings of the Fourth European Academy ofForensic Science Meeting, Helsinki, 2006.

[10] S.E. Bell, L.A. Fido, S.J. Speers, W.J. Armstrong, S. Spratt, Forensic analysis ofarchitectural finishes using fourier transform infrared and raman spectroscopy,Part i: the resin bases, Appl. Spectrosc. 59 (11) (2005) 1333–1339.

[11] S.E. Bell, L.A. Fido, S.J. Speers, W.J. Armstrong, S. Spratt, Forensic analysis ofarchitectural finishes using fourier transform infrared and raman spectroscopy,Part ii: white paint, Appl. Spectrosc. 59 (11) (2005) 1340–1346.

[12] K.W. Smalldon, A.C. Moffat, The calculation of discriminating power for a series ofcorrelated attributes, Sci. Justice 13 (4) (1973) 291–295.

[13] C. Muehlethaler, G. Massonnet, P. Esseiva, The application of chemometrics oninfrared and raman spectra as a tool for the forensic analysis of paints, ForensicSci. Int. 209 (2011) 173–182.

[14] J. Diel, Wet Color Control – An Efficient Concept for Paint Processing, Paint &Coatings Industry, 2004 Available from: http://www.pcimag.com/Articles/Fea-ture_Article/ [updated 2004].


























http://www.pcimag.com/Articles/Feature_Article/

http://www.pcimag.com/Articles/Feature_Article/

survey on batch-to-batch variation in spray paints: a collaborative study

Documents