- Email: [email protected]
Influence of the shaking time on the forensic analysis of FTIR and Raman spectra of spray paints
Forensic Science International 237 (2014) 78–85
Contents lists available at ScienceDirect
Forensic Science International journal homepage: www.elsevier.com/locate/forsciint
Influence of the shaking time on the forensic analysis of FTIR and Raman spectra of spray paints Cyril Muehlethaler a,*, Genevie`ve Massonnet a, Patrick Buzzini b a b
Ecole des Sciences Criminelles, Institut de Police Scientifique, Universite´ de Lausanne, 1015 Lausanne-Dorigny, Switzerland Forensic and Investigative Science, West Virginia University, Morgantown, United States
A R T I C L E I N F O
A B S T R A C T
Article history: Received 7 November 2013 Accepted 29 January 2014 Available online 12 February 2014
In order to decide if replicated measurements of a trace fall within the intra-variability expected for reference paint samples, a forensic scientist has to understand and integrate all reasonable sources of variation. The origins of such variation in spectra can be various, but mainly include differences in components distribution (homogeneity of spraying) or differences originating from the manufacturing process (production batches). Instrumental variation can also be problematic for non-successive measurements. Infrared and Raman spectra were collected to study the homogeneity of the paint distribution after shaking a spray can for times of 0, 1, 2, 3, 4 and 5 min. The results confirm that differences arise in both the spectroscopic techniques used in this study. Mainly, this survey shows that the problematic of shaking is particularly important when the pigment content can be detected from spray paint samples within the infrared domain. In these situations, the signal from the pigment might produce strong absorptions that vary with shaking time, leading to differences in relative intensities with respect to those attributed to the binder. For Raman spectroscopy, it has been shown that a gradient of pigment concentration is observable in some samples depending on the shaking time. The proportion of the signal due to the pigment increases with shaking times from 0 to 1 min and diminishes afterwards, to finally reach stabilization around 3 min of shaking. Not all samples are affected by these differences and it should always be evaluated on a case-by-case basis. From a statistical point-of-view, principal component analyses of the replicates show that the spectra are reproducible after 3 min of shaking. ß 2014 Elsevier Ireland Ltd. All rights reserved.
Keywords: Forensic Paint Spraying Infrared Raman Pigment PCA
1. Introduction Criminal cases involving spray paints are various and in most cases remain linked to vandalism. Generally a trace is recovered, analyzed and compared to paint reference samples such as spray cans belonging to a suspect. The analysis of spray paints characteristics has been deeply investigated. Among the published documents, some papers focused primarily on population studies and discrimination by color type, such as black [1–3], red [4], green [5] or various graffitis paints [6,7]. Other surveys concentrated on paint traces [8,9] and the transfer of microscopic paint droplets on clothes [10,11], as well as paint homogeneity by shaking [12], or discrimination of batches of spray paints [13]. Beside these studies, the question of explaining the origin of small variations and the level of discrimination we can reach has little been addressed. Among these limitations, a large part is
* Corresponding author. Tel.: +41 21 692 46 28; fax: +41 21 692 46 05. E-mail address: [email protected] (C. Muehlethaler). http://dx.doi.org/10.1016/j.forsciint.2014.01.024 0379-0738/ß 2014 Elsevier Ireland Ltd. All rights reserved.
concerned with the interpretation of spectroscopic data (e.g., Infrared, Raman), mainly in deciding how to consider small differences in relative intensities of peaks of paints from two different sources. Statistical methods are certainly able to identify these differences, but the expert has then to decide if these are real differences and how to interpret them [14]. The small differences in spectra can be various, and from different origins when we consider both reference paints or paint traces. For reference paints (e.g., spray cans), when measured in optimal conditions, variations mainly include components distribution (homogeneity of spraying) or differences originating from the manufacturing process (production batches). The use of different instruments is another important source of the spectral variation, given that data acquisition is laboratory/instrumentalas well as procedure/manipulator- dependent. Although normally kept to a minimum by the operator, the instrumental variation can be problematic for non-successive measurements. For paint traces (e.g., graffitis), numerous other factors add to the unwanted variation in spectra, such as contamination from the support, weathering of the surface, or quantity of material collected.
C. Muehlethaler et al. / Forensic Science International 237 (2014) 78–85
The shaking of spray cans was first studied by Zeichner et al. [12], they showed that solid materials were sedimenting at the bottom of the can. In cases of insufficient shaking, these heavier materials were sprayed first due to the proximity with the tube, then followed by solvent mostly that filled the crater formed in the solid materials. Although results were obtained with X-ray analyses, the comparison with spectroscopic techniques such as Raman and FTIR should be addressed, as differences in spectra are expected to be observed as well. All together these considerations may lead to significant changes in the spectral properties of a paint, as the homogeneity is of critical importance for comparing two forensic samples and producing quality reference materials. The present article studies the effect of shaking on the homogeneity and reproducibility of components distribution in the final paint layer for reference spray paints. It compares the results obtained by Zeichner et al. [12] with FTIR and Raman measurements, and presents a statistical point-of-view of the homogeneity relative to the time of shaking using principal component analyses (PCA).
79
either 785 nm HPNIR Renishaw (8–12 mW power on sample) for the Red samples or 633 nm HeNe Renishaw (0.3–0.4 mW power on sample) for the Blue and Green samples. The other tested laser lines gave too much fluorescence and a weak Raman scattering signal. Measurements are performed in situ on dried paint on glass slides. The parameters were optimized depending on the laser, but at least : 2000–200 cm 1 range, 10 s integration time, 1 accumulation. 2.4. Data treatment Statistical treatments were performed using Unscrambler (Camo software), AnalyzeIt MVP module for KnowItAll (Bio-Rad Laboratories) and Matlab (Mathworks). Pretreatments were systematically applied to all spectra, using a standard normal variate (SNV) + detrending (2nd order polynomial) correction. The variables were selected using two intervals for the infrared spectra (600–1850 cm 1 + 2700–3800 cm 1) in order to eliminate uninformative variables. No variable selections were made for the Raman spectra (300–2000 cm 1). Principal component analyses were systematically cross-validated and data mean centered.
2. Materials and methods 3. Results 2.1. Samples 3.1. Fourier transform infrared spectroscopy The sample set is composed of paints of three different colors (red RAL 3000, green RAL 6002, and blue RAL 5010) and three different brands, for a total of nine spray cans. Table 1 regroups the general information about the samples and their composition as found by infrared and Raman analyses detailed below. 2.2. Measurements Paint samples were prepared by spraying onto clean microscopic glass slides. All nine samples were sprayed after 0, 1, 2, 3, 4 and 5 min of shaking using an IKAß laboratory shaker. Three different slides were prepared for each of the above conditions. After spraying, samples were allowed to dry for 48 h. Non-shaken preparations (0 min) were made after a 24 h rest of the can. Five measurements over the surface of each glass slide were made, for a total of 15 replicates for each sample. 2.3. Techniques Infrared measurements were made on a Digilab Excalibur series FTS300 equipped with a UMA600 Microscope, a 15 objective, a mercury cadmium telluride (MCT) detector and Resolutions ProTM software. Sampling method was transmittance on KBr pellets. The measurement parameters were as follow : 150 150 mm window size, 4.0 cm 1 resolution, 32 co-added spectra, and 4000–650cm 1 range. Raman measurements were made on a Renishaw RM1000 with a DML Leica microscope, and 50 objective. The laser sources were Table 1 Description and general information of the sample set. [ACR: acrylic, ALK OPH: orthophthalic alkyd] [C.I. Pigments Names: PR112 (Naphthol red), PB15 (Phthalocyanine blue), PG7 (Phthalocyanine green), PY74 (Mono azo yellow)]. Shop
Brand
Color
Code
Composition
Jumbo Jumbo Jumbo Obi Obi Obi Migros Migros Migros
Eurocolor Eurocolor Eurocolor Dupli-color Dupli-color Dupli-color MColor MColor MColor
Red Green Blue Red Green Blue Red Green Blue
JR JG JB OR OG OB MR MG MB
ALK OPH, PR112 ALK OPH, PG7 (PY74) ALK OPH, PB15 ALK OPH, PR112 ALK OPH, PG7 (PY74) ALK OPH, PB15 ACR, PR112 ACR, PG7 (PY74) ACR, PB15
To estimate the extent of variation between unshaken and shaken spray cans, the spectra were first visually examined and their composition determined based on the main absorption peaks. The binder (resin) always generated the major signal in the spectra. Sometimes IR absorption bands from the pigments were observed, which were partially hidden by the binder peaks. These absorptions were observable in the ‘‘fingerprint’’ spectral region from 1300 to 1800 cm 1. The chemical individual assignment of each vibrational absorptions in this region is always difficult due to many overlapping peaks. They have however a bigger interest for comparison, as the organic pigments may display infrared absorptions in this fingerprint region. From the nine samples, only pigments red (PR112) and pigments yellow (PY74) were visible and produced strong enough IR absorption bands that permit their identification. As first hypothesized by Zeichner et al. [12], the pigments should be differently concentrated with varying shaking times. The spectra that display detectable absorptions bands from the pigments are therefore more likely to present differences. Fig. 1 shows the spectra of sample JR after respectively 0, 1, 2, 3, 4, and 5 min of shaking. A decrease in intensity of the bands attributed to pigments was observed. The proportion with the binder is high at 0 min, and then diminishes and stabilizes after approximately 1– 2 min to remain constant afterwards. For the other samples, either there were no signals from the part of the pigments that could be detected, or the spectra had no varying intensities over the time of shaking as illustrated for sample JB in Fig. 2. These observations confirm that paints are complex mixtures of different compounds and their homogeneity on the painted substrate is influenced by the time of shaking. In order to visualize this inhomogeneity of replicated measurements respective of the shaking time, a PCA was conducted for each sample individually, on data previously mean centered and normalized (Z-scores). All replicates from each shaking time (0, 1, 2, 3, 4 and 5 min) were fed into the statistical model and their relative distances between classes centroids computed using Mahanalobis distances. Fig. 3 presents the results for all the samples. The sample MR presents a special pattern, as all replicates from 1 min fall in a clearly separated group. They have in fact a stronger pigment signal related to the pigment concentration which is significantly higher than other times of shaking. For the other
[(Fig._1)TD$IG]
80
C. Muehlethaler et al. / Forensic Science International 237 (2014) 78–85
Fig. 1. Infrared spectra of sample JR after respectively 0, 1, 2, 3, 4, and 5 min of shaking. Reference of a Pigment Red 112 (Naphthol Red) is illustrated at the bottom. The spectrum at 0 min demonstrates that peaks corresponding to the pigment PR112 are sensibly higher and then decrease with longer shaking times.
samples, it is also observed that the groups corresponding to the times of shaking vary both in terms of spread (homogeneity of the replicated measurements) and separation (distance between the different classes). An observation of the loading plots also confirms that relative differences between the pigments and the binder absorptions are mostly responsible of these separations or inhomogeneity between replicates. As it was witnessed that groups of 3, 4, and 5 min were often closer than others, the Mahanalobis distances between all classes centroids were estimated by boxplot (median and standard deviation) for all samples together (Fig. 4). The ideal situation would be a mean of zero with the smallest deviation, indicating that the two classes cannot be differentiated (all replicates fall on top of each other in the principal component analysis). As Fig. 4 illustrates, this ideal situation is achieved with times of shaking superior to 3 min (e.g., between classes 3–4, 3–5 and 4–5). 3.2. Raman spectroscopy The same procedure was applied to Raman spectra of the nine samples. The Raman spectra collected over the covered range (200–2000 cm 1) are essentially due to the pigment(s) contribution(s). Semi-quantitative differences in peaks intensities with
[(Fig._2)TD$IG]
binder or extenders are not expected. These conditions are proposed for standard Raman measurements [15]. Although the possible extension of the spectral range to 4000 cm 1 in order to offer the potential of observing CH bands from the binder, the choice to restrict to 200–2000 cm 1 was made to avoid the large fluorescence and baselines shifts sometimes arising after 2500 cm 1. Fig. 5 presents the Raman spectra of samples JG after 0, 1, 2, 3, 4 and 5 min of shaking. The results show that Raman spectra of the paint samples do not have variations in relative intensities of the peaks. All spectra are similar and very reproducible independently of the shaking time. Although no relative intensities differences with the binder are expected, the concentration of the pigments should nonetheless be changing with the shaking times. Observation of the spectra on a common scale confirms that higher intensities (in counts) are associated with smaller shaking times for samples measured under the same conditions. For sample MR it is particularly evident that 1 min shaking time produced absolute counts almost 2–3 times higher than those obtained with 3, 4, and 5 min shaking times (Fig. 6). This example also demonstrates that the concentration of the pigment increases from 0 to 1 min and then decreases to stabilize after approximately 2–3 min, where this effect is then barely visible. Although particularly significant for the sample MR,
Fig. 2. Infrared spectra of sample JB after respectively 0, 1, 2, 3, 4, and 5 min of shaking. Pigment Blue 15 (as identified by Raman spectroscopy) do not produce characteristic infrared absorption in the samples spectra. The spectra are stable considering the time of shaking and represent essentially the binder composition.
[(Fig._3)TD$IG]
C. Muehlethaler et al. / Forensic Science International 237 (2014) 78–85
Fig. 3. Principal component analysis (PCA) of the Infrared spectra of all the samples. Colored classes represent different times of shaking.
[(Fig._4)TD$IG]
Fig. 4. Boxplot representing relative distances between groups centroids for the infrared spectra (Mahalanobis distances).
81
[(Fig._5)TD$IG]
82
C. Muehlethaler et al. / Forensic Science International 237 (2014) 78–85
Fig. 5. Raman spectra of sample JG after 0, 1, 2, 3, 4 and 5 min of shaking (from top to bottom) with a reference spectrum of Pigment Green 7 (bottom).
this effect was not observed on every sample and exceptions were present with equivalent counts for all shaking times. Unlike FTIR however, it was observed that the most important source of variation in the recorded spectra was the baseline, which contributed significantly to the differences in PCA score-plots and loadings (Fig. 7). The samples had varying degrees of fluorescence and sometimes baseline variation was quite significant. The same
[(Fig._6)TD$IG]
procedure for calculating the distances between classes in PCA scores-plots was used as for FTIR. A highest separation is visible at 0 min of shaking for almost all the samples (Fig. 8). Afterwards all distributions present particularities such as inhomogeneity or separation between classes. The produced boxplots show that at higher shaking times, the Raman values are more variable than FTIR measurements – they were closer to a value 0 of mean and
Fig. 6. Raman spectra of the sample MR after 0, 1, 2, 3, 4 and 5 min of shaking (from bottom to top), represented using a common scale. Maximum counts calculated for the peak at 1358 cm 1 for the 15 replicates of each condition are plotted using boxplots. Highest values are present for the spectrum at 1 min, confirming a higher pigment concentration at the sample location.
[(Fig._7)TD$IG]
C. Muehlethaler et al. / Forensic Science International 237 (2014) 78–85
83
Fig. 7. Principal component analysis of the Raman spectra of all the samples. Colored classes represent different times of shaking.
[(Fig._8)TD$IG]
lower standard deviations, which is not the case here. This represents however the normal variation expected in Raman spectra, which are less reproducible and more subjected to instrumental variation than infrared analyses. This aspect mainly
reflects in varying baselines that could be corrected using adapted pretreatments. It is also assumed that Raman analyses are very sensitive to the surface’s aspect (e.g., flat, porous, polished) and the following ability to focus the laser spot into the microscope [16].
Fig. 8. Boxplot representing relative distances between groups centroids for the Raman spectra (Mahalanobis distances).
84
C. Muehlethaler et al. / Forensic Science International 237 (2014) 78–85
As compared to Infrared spectra that require times of shaking superior to 3 min, the boxplot of Raman analyses shows that the recorded spectra are already constant after only a 1 min shaking time. This makes sense given that Raman spectroscopy is sensitive to colorants and very few to none scattering signals from the binders are recorded within the studied spectral range (200–2000 cm 1). The complex mixture of binder and extenders has no effects on Raman spectra and the pigment content only is important. The effect of the pigment concentration can be observed by considering the absolute counts (Raman peak intensity) but as spectra are generally compared using normalized values this effect is minimized. For mixture of pigments however, the proportions between the two individual pigments can possibly vary if insufficient shaking is applied. This aspect was however not observed in this study for the green spray samples containing the two identified pigments PG 7 and PY 74. The reason was mainly that the two pigments had different affinities with the used laser lines. PG 7 was detected using the lasers at 633 nm or 785 nm (Fig. 5), while PY 74 could be identified using the 514 nm laser line but also by FTIR spectroscopy. The fact that they are sensitive to different wavelengths renders their confusion in one single spectrum less likely. 4. Discussion 4.1. Influence of the shaking time The results confirm the explanation given by Zeichner et al. [12] and also that changes are detected with the spectroscopic techniques utilized in this study. The initial explanation was that ‘‘at the beginning of spraying unshaken spray paints, a high concentration of inorganic material comes out first, since it is in the bottom of the can where the tube is located. Following that, the solvent above the settled solids ‘‘cleans itself a way’’ into the tube by creating a crater around the end of the tube so that, afterward, mostly solvent comes out during spraying’’ [12]. Their initial study, being mainly based on X-ray analyses, considered the inorganic constituents (including inorganic fillers and pigments) as the settled solids. In the present study, organic pigments were considered as well. However, not all of these organic or inorganic materials are necessarily visible in FTIR or Raman analyses. They can also be present (hence reducing the proportion of binder) without producing detectable signals and clearly defined peaks. Two different situations need to be discussed. First, when the cans were not shaken at all, it can be considered that all the pigments are settling in the bottom (Fig. 9). A higher pigment concentration will be initially sprayed followed by the solvent and the binder. This situation can lead to pigment concentration very
[(Fig._9)TD$IG]
Fig. 9. Schematic illustration of the bottom of a spray paint can. The proximity of settled materials with the tube explains their differential concentration during spraying; (a) after 24 h rest: pigments are in the bottom and will be sprayed first followed by solvent/binder, (b) 1 or 2 min shaking: pigments are dispersed in solution still being more concentrated at the bottom, (c) more than 3 min of shaking: pigments are homogeneously distributed in the solvent/binder mixture.
high at some spots and then relatively low afterwards during a continuous spraying process. For small paintings such as glass slides, this irregular spraying over time was not observed, even between the three slides (less than 10 s). It is however conceivable to observe this situation for larger surfaces such as graffitis or wall/ objects paintings, as it was initially observed for times of paintings between 10 and 25 s [12]. The second case refers to situations when the cans were shaken. When the shaking is insufficient (e.g., 1 or 2 min), pigments are dispersed in the binder, but their heavier weight and/or binder (and solvent) viscosity retains them toward the bottom side of the can, where their proportion remains higher. Compared to unshaken cans the spraying will therefore be more constant over time although pigments will be over-concentrated. This situation was noted for sample MR (Fig. 6), where Raman spectra obtained with 1 min of shaking generated bands almost twice more intense than those obtained with other shaking conditions. To ensure (a) correct concentration of the pigments versus binder and (b) constant spraying proportions, a shaking time of 3 min at least should be respected when preparing reference materials. Although valid in theory, these considerations are in practice not so straightforward to apply. Depending on the type of paint or the type of spray can, the pigments signal may not vary significantly. The differences being very minor, they can be masked by sample preparation or measurements conditions (e.g., sample thickness, resolution, scans number), affecting both spectral quality and overall absorbance. Concerning the pigment(s) identification using FTIR solely, it seems difficult to formulate conclusions regarding the shaking time as it seems too dependent on the technique. However by considering the whole spectrum and composition (binder, extender, pigments), the statistical results support that a shaking time of 3 min at least is necessary to ensure optimal homogeneity of the painted surface. 4.2. Differences between FTIR and Raman It was also observed that Infrared and Raman spectroscopic methods are variably sensitive to these differences of shaking. The Raman technique, for example, is not sensitive to differences of proportions between pigments and binders over the studied spectral range (200–2000 cm 1), but they are affected solely by the pigments concentration. This will be observable when compared in absolute intensities, but barely visible if normalized spectra are used. FTIR spectroscopy, however, is very sensitive to these differences of concentrations, as it measures both absorptions of the binder and (but not always) of the pigments. The spectra are complex combinations of both of these signals. This is of significant importance because certain pigments do not present strong IR absorption characteristic bands making their signal less likely to be detected, even at shorter shaking times. In that case, differences due to shaking tend to diminish. The production of representative reference materials for paint analyses should therefore be strictly controlled in practice. For traces on the other hand, it seems difficult to ascertain that the author rigorously respected 3 min of shaking. What would be possible is therefore to observe a slightly higher pigment concentration in infrared spectra of the paint traces (situation corresponding to 0, 1 or 2 min of shaking). The very high diversity of paint compositions makes a generalization of these findings difficult. The shaking influence depends on binders, extenders and pigments composition, but also on other factors such as the level of filling of the can, viscosity, solvent type or kind of propellant gases. Furthermore, considering paint traces, the effect is more pronounced for some paints, while others are not affected at all by the shaking. As a general recommendation for handling a situation with possible differences
C. Muehlethaler et al. / Forensic Science International 237 (2014) 78–85
due to the shaking it is necessary to adapt and consider them on a case-by-case basis. The first necessity is to produce reference paint samples after 3 min of shaking, possibly followed with new reference samples at 0 min. The samples at 3 min will permit to estimate the homogeneity of the surface and the range of variation expected for the unknown/trace samples (e.g., graffiti), while samples at 0 min will permit to account for the occurrence of a pigment differential concentration. 5. Conclusion The potential differences in spectral properties arising from insufficient shaking of spray paint samples were studied. It has been shown that shaking time is important to consider, particularly when pigments presenting characteristics infrared absorptions are present. In that case, the pigments sometimes present a stronger signal during the early stages of shaking (0 min– 1min), then diminishing gradually with longer times (2 min– 5 min). The other paints are more stable and produce similar spectra at every time. Concerning Raman spectroscopy, the effect of pigment concentration is noticeable for some of the samples using absolute intensities counts. For some samples, a gradient of concentration was observed, with an increase in pigment concentration from 0 to 1 min followed by a decrease and stabilization around 3 min of shaking. It has also been shown that from a statistical point-of-view, the spectra are best reproducible after at least 3 min of shaking. Under these conditions an appropriate number of replicates is able accounting for the observed heterogeneity of the painted surface. References [1] R. Gosse, S. Milet, B. Espanet, Discrimination of black spray paints, in: Proceedings of the European Paint and Glass Group Meeting, Berlin, 2005.
85
[2] S. Ryland, Discrimination of retail black spray paints, J. Am. Soc. Trace Evidence Examiners 1 (2) (2010) 109–126. [3] F. Govaert, G. Roy, B. Decruyenaere, D. Ziernicki, Analysis of black spray paints by Fourier transform infrared spectrometry, X-ray fluorescence and visible microscopy, Prob. Forensic Sci. XLVII (2001) 333–339. [4] 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. [5] P. Buzzini, G. Massonnet, A market study of green spray paints by Fourier transform infrared (FTIR) and Raman spectroscopy, Sci. Justice 44 (3) (2004) 123–131. [6] L. Marsh, Some call it art: Case studies investigating the spraying of illegal graffiti in the UK, in: Proceedings of the Trace Evidence Symposium, Clearwater Beach, Florida, 2007, http://projects.nfstc.org/trace/index2007.htm. [7] K.-M. Pun, P. Buzzini, P. Margot, Cre´ation d‘une banque de donne´es de graffitis a` des fins d‘investigation: e´tude pre´liminaire, Rev. Int. Criminol. Police Techn. Sci. 2 (2006) 215–232. [8] J. Zieba-Palus, Examination of spray paints by the use of reflection technique of microinfrared spectroscopy, J. Mol. Struct. 744–747 (2005) 229–234. [9] J.M. Milczarek, J. Zieba-Palus, Examination of spray paints on plasters by the use of pyrolysis-gas chromatography/mass spectrometry for forensic purposes, J. Anal. Appl. Pyrolysis 86 (2) (2009) 252–259. [10] C.D.J. Krausher, Characteristics of aerosol paint transfer and dispersal, Can. Soc. Forensic Sci. J. 27 (3) (1994) 125–142. [11] D. Marin, N. Berger, P. Buzzini, G. Massonnet, Transfer, detection and in situ raman analysis of spray paint traces on clothes, in: Proceedings of the 10th European Paint and Glass Group Meeting, Prague, 2004. [12] A. Zeichner, N. Levin, E. Landau, A study of paint coat characteristics produced by spray paints from shaken and nonshaken spray cans, J. Forensic Sci. 37 (2) (1992) 542–555. [13] C. Muehlethaler, G. Massonnet, M. Deviterne, M. Bradley, A. Herrero, I. Diaz de, Lezana, S. Lauper, D. Dubois, J. Geyer-Lippmann, S. Ketterer, S. Milet, M. Bertrand, W. Langer, B. Plage, G. Gorzawski, V. Lamothe, L. Marsh, R. Turunen, Survey on batch-to-batch variation in spray paints: a collaborative study, Forensic Sci. Int. 229 (2013) 80–91. [14] C. Muehlethaler, G. Massonnet, P. Esseiva, The application of chemometrics on infrared and raman spectra as a tool for the forensic analysis of paints, Forensic Sci. Int. 209 (2011) 173–182. [15] E.N.F.S.I. European Paint, Glass Group, Manual of best practice for the forensic examination of paint. Appendix E, Raman Spectrosc. (2009). [16] D. Lambert, C. Muehlethaler, L. Gueissaz, G. Massonnet, Raman analysis of multilayer automotive paints in forensic science: measurement variability and depth profile, in: Proceedings of the 7th International Congress on the Application of Raman Spectroscopy in Art and Archaeology, Ljubljana, 2013.
Contents lists available at ScienceDirect
Forensic Science International journal homepage: www.elsevier.com/locate/forsciint
Influence of the shaking time on the forensic analysis of FTIR and Raman spectra of spray paints Cyril Muehlethaler a,*, Genevie`ve Massonnet a, Patrick Buzzini b a b
Ecole des Sciences Criminelles, Institut de Police Scientifique, Universite´ de Lausanne, 1015 Lausanne-Dorigny, Switzerland Forensic and Investigative Science, West Virginia University, Morgantown, United States
A R T I C L E I N F O
A B S T R A C T
Article history: Received 7 November 2013 Accepted 29 January 2014 Available online 12 February 2014
In order to decide if replicated measurements of a trace fall within the intra-variability expected for reference paint samples, a forensic scientist has to understand and integrate all reasonable sources of variation. The origins of such variation in spectra can be various, but mainly include differences in components distribution (homogeneity of spraying) or differences originating from the manufacturing process (production batches). Instrumental variation can also be problematic for non-successive measurements. Infrared and Raman spectra were collected to study the homogeneity of the paint distribution after shaking a spray can for times of 0, 1, 2, 3, 4 and 5 min. The results confirm that differences arise in both the spectroscopic techniques used in this study. Mainly, this survey shows that the problematic of shaking is particularly important when the pigment content can be detected from spray paint samples within the infrared domain. In these situations, the signal from the pigment might produce strong absorptions that vary with shaking time, leading to differences in relative intensities with respect to those attributed to the binder. For Raman spectroscopy, it has been shown that a gradient of pigment concentration is observable in some samples depending on the shaking time. The proportion of the signal due to the pigment increases with shaking times from 0 to 1 min and diminishes afterwards, to finally reach stabilization around 3 min of shaking. Not all samples are affected by these differences and it should always be evaluated on a case-by-case basis. From a statistical point-of-view, principal component analyses of the replicates show that the spectra are reproducible after 3 min of shaking. ß 2014 Elsevier Ireland Ltd. All rights reserved.
Keywords: Forensic Paint Spraying Infrared Raman Pigment PCA
1. Introduction Criminal cases involving spray paints are various and in most cases remain linked to vandalism. Generally a trace is recovered, analyzed and compared to paint reference samples such as spray cans belonging to a suspect. The analysis of spray paints characteristics has been deeply investigated. Among the published documents, some papers focused primarily on population studies and discrimination by color type, such as black [1–3], red [4], green [5] or various graffitis paints [6,7]. Other surveys concentrated on paint traces [8,9] and the transfer of microscopic paint droplets on clothes [10,11], as well as paint homogeneity by shaking [12], or discrimination of batches of spray paints [13]. Beside these studies, the question of explaining the origin of small variations and the level of discrimination we can reach has little been addressed. Among these limitations, a large part is
* Corresponding author. Tel.: +41 21 692 46 28; fax: +41 21 692 46 05. E-mail address: [email protected] (C. Muehlethaler). http://dx.doi.org/10.1016/j.forsciint.2014.01.024 0379-0738/ß 2014 Elsevier Ireland Ltd. All rights reserved.
concerned with the interpretation of spectroscopic data (e.g., Infrared, Raman), mainly in deciding how to consider small differences in relative intensities of peaks of paints from two different sources. Statistical methods are certainly able to identify these differences, but the expert has then to decide if these are real differences and how to interpret them [14]. The small differences in spectra can be various, and from different origins when we consider both reference paints or paint traces. For reference paints (e.g., spray cans), when measured in optimal conditions, variations mainly include components distribution (homogeneity of spraying) or differences originating from the manufacturing process (production batches). The use of different instruments is another important source of the spectral variation, given that data acquisition is laboratory/instrumentalas well as procedure/manipulator- dependent. Although normally kept to a minimum by the operator, the instrumental variation can be problematic for non-successive measurements. For paint traces (e.g., graffitis), numerous other factors add to the unwanted variation in spectra, such as contamination from the support, weathering of the surface, or quantity of material collected.
C. Muehlethaler et al. / Forensic Science International 237 (2014) 78–85
The shaking of spray cans was first studied by Zeichner et al. [12], they showed that solid materials were sedimenting at the bottom of the can. In cases of insufficient shaking, these heavier materials were sprayed first due to the proximity with the tube, then followed by solvent mostly that filled the crater formed in the solid materials. Although results were obtained with X-ray analyses, the comparison with spectroscopic techniques such as Raman and FTIR should be addressed, as differences in spectra are expected to be observed as well. All together these considerations may lead to significant changes in the spectral properties of a paint, as the homogeneity is of critical importance for comparing two forensic samples and producing quality reference materials. The present article studies the effect of shaking on the homogeneity and reproducibility of components distribution in the final paint layer for reference spray paints. It compares the results obtained by Zeichner et al. [12] with FTIR and Raman measurements, and presents a statistical point-of-view of the homogeneity relative to the time of shaking using principal component analyses (PCA).
79
either 785 nm HPNIR Renishaw (8–12 mW power on sample) for the Red samples or 633 nm HeNe Renishaw (0.3–0.4 mW power on sample) for the Blue and Green samples. The other tested laser lines gave too much fluorescence and a weak Raman scattering signal. Measurements are performed in situ on dried paint on glass slides. The parameters were optimized depending on the laser, but at least : 2000–200 cm 1 range, 10 s integration time, 1 accumulation. 2.4. Data treatment Statistical treatments were performed using Unscrambler (Camo software), AnalyzeIt MVP module for KnowItAll (Bio-Rad Laboratories) and Matlab (Mathworks). Pretreatments were systematically applied to all spectra, using a standard normal variate (SNV) + detrending (2nd order polynomial) correction. The variables were selected using two intervals for the infrared spectra (600–1850 cm 1 + 2700–3800 cm 1) in order to eliminate uninformative variables. No variable selections were made for the Raman spectra (300–2000 cm 1). Principal component analyses were systematically cross-validated and data mean centered.
2. Materials and methods 3. Results 2.1. Samples 3.1. Fourier transform infrared spectroscopy The sample set is composed of paints of three different colors (red RAL 3000, green RAL 6002, and blue RAL 5010) and three different brands, for a total of nine spray cans. Table 1 regroups the general information about the samples and their composition as found by infrared and Raman analyses detailed below. 2.2. Measurements Paint samples were prepared by spraying onto clean microscopic glass slides. All nine samples were sprayed after 0, 1, 2, 3, 4 and 5 min of shaking using an IKAß laboratory shaker. Three different slides were prepared for each of the above conditions. After spraying, samples were allowed to dry for 48 h. Non-shaken preparations (0 min) were made after a 24 h rest of the can. Five measurements over the surface of each glass slide were made, for a total of 15 replicates for each sample. 2.3. Techniques Infrared measurements were made on a Digilab Excalibur series FTS300 equipped with a UMA600 Microscope, a 15 objective, a mercury cadmium telluride (MCT) detector and Resolutions ProTM software. Sampling method was transmittance on KBr pellets. The measurement parameters were as follow : 150 150 mm window size, 4.0 cm 1 resolution, 32 co-added spectra, and 4000–650cm 1 range. Raman measurements were made on a Renishaw RM1000 with a DML Leica microscope, and 50 objective. The laser sources were Table 1 Description and general information of the sample set. [ACR: acrylic, ALK OPH: orthophthalic alkyd] [C.I. Pigments Names: PR112 (Naphthol red), PB15 (Phthalocyanine blue), PG7 (Phthalocyanine green), PY74 (Mono azo yellow)]. Shop
Brand
Color
Code
Composition
Jumbo Jumbo Jumbo Obi Obi Obi Migros Migros Migros
Eurocolor Eurocolor Eurocolor Dupli-color Dupli-color Dupli-color MColor MColor MColor
Red Green Blue Red Green Blue Red Green Blue
JR JG JB OR OG OB MR MG MB
ALK OPH, PR112 ALK OPH, PG7 (PY74) ALK OPH, PB15 ALK OPH, PR112 ALK OPH, PG7 (PY74) ALK OPH, PB15 ACR, PR112 ACR, PG7 (PY74) ACR, PB15
To estimate the extent of variation between unshaken and shaken spray cans, the spectra were first visually examined and their composition determined based on the main absorption peaks. The binder (resin) always generated the major signal in the spectra. Sometimes IR absorption bands from the pigments were observed, which were partially hidden by the binder peaks. These absorptions were observable in the ‘‘fingerprint’’ spectral region from 1300 to 1800 cm 1. The chemical individual assignment of each vibrational absorptions in this region is always difficult due to many overlapping peaks. They have however a bigger interest for comparison, as the organic pigments may display infrared absorptions in this fingerprint region. From the nine samples, only pigments red (PR112) and pigments yellow (PY74) were visible and produced strong enough IR absorption bands that permit their identification. As first hypothesized by Zeichner et al. [12], the pigments should be differently concentrated with varying shaking times. The spectra that display detectable absorptions bands from the pigments are therefore more likely to present differences. Fig. 1 shows the spectra of sample JR after respectively 0, 1, 2, 3, 4, and 5 min of shaking. A decrease in intensity of the bands attributed to pigments was observed. The proportion with the binder is high at 0 min, and then diminishes and stabilizes after approximately 1– 2 min to remain constant afterwards. For the other samples, either there were no signals from the part of the pigments that could be detected, or the spectra had no varying intensities over the time of shaking as illustrated for sample JB in Fig. 2. These observations confirm that paints are complex mixtures of different compounds and their homogeneity on the painted substrate is influenced by the time of shaking. In order to visualize this inhomogeneity of replicated measurements respective of the shaking time, a PCA was conducted for each sample individually, on data previously mean centered and normalized (Z-scores). All replicates from each shaking time (0, 1, 2, 3, 4 and 5 min) were fed into the statistical model and their relative distances between classes centroids computed using Mahanalobis distances. Fig. 3 presents the results for all the samples. The sample MR presents a special pattern, as all replicates from 1 min fall in a clearly separated group. They have in fact a stronger pigment signal related to the pigment concentration which is significantly higher than other times of shaking. For the other
[(Fig._1)TD$IG]
80
C. Muehlethaler et al. / Forensic Science International 237 (2014) 78–85
Fig. 1. Infrared spectra of sample JR after respectively 0, 1, 2, 3, 4, and 5 min of shaking. Reference of a Pigment Red 112 (Naphthol Red) is illustrated at the bottom. The spectrum at 0 min demonstrates that peaks corresponding to the pigment PR112 are sensibly higher and then decrease with longer shaking times.
samples, it is also observed that the groups corresponding to the times of shaking vary both in terms of spread (homogeneity of the replicated measurements) and separation (distance between the different classes). An observation of the loading plots also confirms that relative differences between the pigments and the binder absorptions are mostly responsible of these separations or inhomogeneity between replicates. As it was witnessed that groups of 3, 4, and 5 min were often closer than others, the Mahanalobis distances between all classes centroids were estimated by boxplot (median and standard deviation) for all samples together (Fig. 4). The ideal situation would be a mean of zero with the smallest deviation, indicating that the two classes cannot be differentiated (all replicates fall on top of each other in the principal component analysis). As Fig. 4 illustrates, this ideal situation is achieved with times of shaking superior to 3 min (e.g., between classes 3–4, 3–5 and 4–5). 3.2. Raman spectroscopy The same procedure was applied to Raman spectra of the nine samples. The Raman spectra collected over the covered range (200–2000 cm 1) are essentially due to the pigment(s) contribution(s). Semi-quantitative differences in peaks intensities with
[(Fig._2)TD$IG]
binder or extenders are not expected. These conditions are proposed for standard Raman measurements [15]. Although the possible extension of the spectral range to 4000 cm 1 in order to offer the potential of observing CH bands from the binder, the choice to restrict to 200–2000 cm 1 was made to avoid the large fluorescence and baselines shifts sometimes arising after 2500 cm 1. Fig. 5 presents the Raman spectra of samples JG after 0, 1, 2, 3, 4 and 5 min of shaking. The results show that Raman spectra of the paint samples do not have variations in relative intensities of the peaks. All spectra are similar and very reproducible independently of the shaking time. Although no relative intensities differences with the binder are expected, the concentration of the pigments should nonetheless be changing with the shaking times. Observation of the spectra on a common scale confirms that higher intensities (in counts) are associated with smaller shaking times for samples measured under the same conditions. For sample MR it is particularly evident that 1 min shaking time produced absolute counts almost 2–3 times higher than those obtained with 3, 4, and 5 min shaking times (Fig. 6). This example also demonstrates that the concentration of the pigment increases from 0 to 1 min and then decreases to stabilize after approximately 2–3 min, where this effect is then barely visible. Although particularly significant for the sample MR,
Fig. 2. Infrared spectra of sample JB after respectively 0, 1, 2, 3, 4, and 5 min of shaking. Pigment Blue 15 (as identified by Raman spectroscopy) do not produce characteristic infrared absorption in the samples spectra. The spectra are stable considering the time of shaking and represent essentially the binder composition.
[(Fig._3)TD$IG]
C. Muehlethaler et al. / Forensic Science International 237 (2014) 78–85
Fig. 3. Principal component analysis (PCA) of the Infrared spectra of all the samples. Colored classes represent different times of shaking.
[(Fig._4)TD$IG]
Fig. 4. Boxplot representing relative distances between groups centroids for the infrared spectra (Mahalanobis distances).
81
[(Fig._5)TD$IG]
82
C. Muehlethaler et al. / Forensic Science International 237 (2014) 78–85
Fig. 5. Raman spectra of sample JG after 0, 1, 2, 3, 4 and 5 min of shaking (from top to bottom) with a reference spectrum of Pigment Green 7 (bottom).
this effect was not observed on every sample and exceptions were present with equivalent counts for all shaking times. Unlike FTIR however, it was observed that the most important source of variation in the recorded spectra was the baseline, which contributed significantly to the differences in PCA score-plots and loadings (Fig. 7). The samples had varying degrees of fluorescence and sometimes baseline variation was quite significant. The same
[(Fig._6)TD$IG]
procedure for calculating the distances between classes in PCA scores-plots was used as for FTIR. A highest separation is visible at 0 min of shaking for almost all the samples (Fig. 8). Afterwards all distributions present particularities such as inhomogeneity or separation between classes. The produced boxplots show that at higher shaking times, the Raman values are more variable than FTIR measurements – they were closer to a value 0 of mean and
Fig. 6. Raman spectra of the sample MR after 0, 1, 2, 3, 4 and 5 min of shaking (from bottom to top), represented using a common scale. Maximum counts calculated for the peak at 1358 cm 1 for the 15 replicates of each condition are plotted using boxplots. Highest values are present for the spectrum at 1 min, confirming a higher pigment concentration at the sample location.
[(Fig._7)TD$IG]
C. Muehlethaler et al. / Forensic Science International 237 (2014) 78–85
83
Fig. 7. Principal component analysis of the Raman spectra of all the samples. Colored classes represent different times of shaking.
[(Fig._8)TD$IG]
lower standard deviations, which is not the case here. This represents however the normal variation expected in Raman spectra, which are less reproducible and more subjected to instrumental variation than infrared analyses. This aspect mainly
reflects in varying baselines that could be corrected using adapted pretreatments. It is also assumed that Raman analyses are very sensitive to the surface’s aspect (e.g., flat, porous, polished) and the following ability to focus the laser spot into the microscope [16].
Fig. 8. Boxplot representing relative distances between groups centroids for the Raman spectra (Mahalanobis distances).
84
C. Muehlethaler et al. / Forensic Science International 237 (2014) 78–85
As compared to Infrared spectra that require times of shaking superior to 3 min, the boxplot of Raman analyses shows that the recorded spectra are already constant after only a 1 min shaking time. This makes sense given that Raman spectroscopy is sensitive to colorants and very few to none scattering signals from the binders are recorded within the studied spectral range (200–2000 cm 1). The complex mixture of binder and extenders has no effects on Raman spectra and the pigment content only is important. The effect of the pigment concentration can be observed by considering the absolute counts (Raman peak intensity) but as spectra are generally compared using normalized values this effect is minimized. For mixture of pigments however, the proportions between the two individual pigments can possibly vary if insufficient shaking is applied. This aspect was however not observed in this study for the green spray samples containing the two identified pigments PG 7 and PY 74. The reason was mainly that the two pigments had different affinities with the used laser lines. PG 7 was detected using the lasers at 633 nm or 785 nm (Fig. 5), while PY 74 could be identified using the 514 nm laser line but also by FTIR spectroscopy. The fact that they are sensitive to different wavelengths renders their confusion in one single spectrum less likely. 4. Discussion 4.1. Influence of the shaking time The results confirm the explanation given by Zeichner et al. [12] and also that changes are detected with the spectroscopic techniques utilized in this study. The initial explanation was that ‘‘at the beginning of spraying unshaken spray paints, a high concentration of inorganic material comes out first, since it is in the bottom of the can where the tube is located. Following that, the solvent above the settled solids ‘‘cleans itself a way’’ into the tube by creating a crater around the end of the tube so that, afterward, mostly solvent comes out during spraying’’ [12]. Their initial study, being mainly based on X-ray analyses, considered the inorganic constituents (including inorganic fillers and pigments) as the settled solids. In the present study, organic pigments were considered as well. However, not all of these organic or inorganic materials are necessarily visible in FTIR or Raman analyses. They can also be present (hence reducing the proportion of binder) without producing detectable signals and clearly defined peaks. Two different situations need to be discussed. First, when the cans were not shaken at all, it can be considered that all the pigments are settling in the bottom (Fig. 9). A higher pigment concentration will be initially sprayed followed by the solvent and the binder. This situation can lead to pigment concentration very
[(Fig._9)TD$IG]
Fig. 9. Schematic illustration of the bottom of a spray paint can. The proximity of settled materials with the tube explains their differential concentration during spraying; (a) after 24 h rest: pigments are in the bottom and will be sprayed first followed by solvent/binder, (b) 1 or 2 min shaking: pigments are dispersed in solution still being more concentrated at the bottom, (c) more than 3 min of shaking: pigments are homogeneously distributed in the solvent/binder mixture.
high at some spots and then relatively low afterwards during a continuous spraying process. For small paintings such as glass slides, this irregular spraying over time was not observed, even between the three slides (less than 10 s). It is however conceivable to observe this situation for larger surfaces such as graffitis or wall/ objects paintings, as it was initially observed for times of paintings between 10 and 25 s [12]. The second case refers to situations when the cans were shaken. When the shaking is insufficient (e.g., 1 or 2 min), pigments are dispersed in the binder, but their heavier weight and/or binder (and solvent) viscosity retains them toward the bottom side of the can, where their proportion remains higher. Compared to unshaken cans the spraying will therefore be more constant over time although pigments will be over-concentrated. This situation was noted for sample MR (Fig. 6), where Raman spectra obtained with 1 min of shaking generated bands almost twice more intense than those obtained with other shaking conditions. To ensure (a) correct concentration of the pigments versus binder and (b) constant spraying proportions, a shaking time of 3 min at least should be respected when preparing reference materials. Although valid in theory, these considerations are in practice not so straightforward to apply. Depending on the type of paint or the type of spray can, the pigments signal may not vary significantly. The differences being very minor, they can be masked by sample preparation or measurements conditions (e.g., sample thickness, resolution, scans number), affecting both spectral quality and overall absorbance. Concerning the pigment(s) identification using FTIR solely, it seems difficult to formulate conclusions regarding the shaking time as it seems too dependent on the technique. However by considering the whole spectrum and composition (binder, extender, pigments), the statistical results support that a shaking time of 3 min at least is necessary to ensure optimal homogeneity of the painted surface. 4.2. Differences between FTIR and Raman It was also observed that Infrared and Raman spectroscopic methods are variably sensitive to these differences of shaking. The Raman technique, for example, is not sensitive to differences of proportions between pigments and binders over the studied spectral range (200–2000 cm 1), but they are affected solely by the pigments concentration. This will be observable when compared in absolute intensities, but barely visible if normalized spectra are used. FTIR spectroscopy, however, is very sensitive to these differences of concentrations, as it measures both absorptions of the binder and (but not always) of the pigments. The spectra are complex combinations of both of these signals. This is of significant importance because certain pigments do not present strong IR absorption characteristic bands making their signal less likely to be detected, even at shorter shaking times. In that case, differences due to shaking tend to diminish. The production of representative reference materials for paint analyses should therefore be strictly controlled in practice. For traces on the other hand, it seems difficult to ascertain that the author rigorously respected 3 min of shaking. What would be possible is therefore to observe a slightly higher pigment concentration in infrared spectra of the paint traces (situation corresponding to 0, 1 or 2 min of shaking). The very high diversity of paint compositions makes a generalization of these findings difficult. The shaking influence depends on binders, extenders and pigments composition, but also on other factors such as the level of filling of the can, viscosity, solvent type or kind of propellant gases. Furthermore, considering paint traces, the effect is more pronounced for some paints, while others are not affected at all by the shaking. As a general recommendation for handling a situation with possible differences
C. Muehlethaler et al. / Forensic Science International 237 (2014) 78–85
due to the shaking it is necessary to adapt and consider them on a case-by-case basis. The first necessity is to produce reference paint samples after 3 min of shaking, possibly followed with new reference samples at 0 min. The samples at 3 min will permit to estimate the homogeneity of the surface and the range of variation expected for the unknown/trace samples (e.g., graffiti), while samples at 0 min will permit to account for the occurrence of a pigment differential concentration. 5. Conclusion The potential differences in spectral properties arising from insufficient shaking of spray paint samples were studied. It has been shown that shaking time is important to consider, particularly when pigments presenting characteristics infrared absorptions are present. In that case, the pigments sometimes present a stronger signal during the early stages of shaking (0 min– 1min), then diminishing gradually with longer times (2 min– 5 min). The other paints are more stable and produce similar spectra at every time. Concerning Raman spectroscopy, the effect of pigment concentration is noticeable for some of the samples using absolute intensities counts. For some samples, a gradient of concentration was observed, with an increase in pigment concentration from 0 to 1 min followed by a decrease and stabilization around 3 min of shaking. It has also been shown that from a statistical point-of-view, the spectra are best reproducible after at least 3 min of shaking. Under these conditions an appropriate number of replicates is able accounting for the observed heterogeneity of the painted surface. References [1] R. Gosse, S. Milet, B. Espanet, Discrimination of black spray paints, in: Proceedings of the European Paint and Glass Group Meeting, Berlin, 2005.
85
[2] S. Ryland, Discrimination of retail black spray paints, J. Am. Soc. Trace Evidence Examiners 1 (2) (2010) 109–126. [3] F. Govaert, G. Roy, B. Decruyenaere, D. Ziernicki, Analysis of black spray paints by Fourier transform infrared spectrometry, X-ray fluorescence and visible microscopy, Prob. Forensic Sci. XLVII (2001) 333–339. [4] 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. [5] P. Buzzini, G. Massonnet, A market study of green spray paints by Fourier transform infrared (FTIR) and Raman spectroscopy, Sci. Justice 44 (3) (2004) 123–131. [6] L. Marsh, Some call it art: Case studies investigating the spraying of illegal graffiti in the UK, in: Proceedings of the Trace Evidence Symposium, Clearwater Beach, Florida, 2007, http://projects.nfstc.org/trace/index2007.htm. [7] K.-M. Pun, P. Buzzini, P. Margot, Cre´ation d‘une banque de donne´es de graffitis a` des fins d‘investigation: e´tude pre´liminaire, Rev. Int. Criminol. Police Techn. Sci. 2 (2006) 215–232. [8] J. Zieba-Palus, Examination of spray paints by the use of reflection technique of microinfrared spectroscopy, J. Mol. Struct. 744–747 (2005) 229–234. [9] J.M. Milczarek, J. Zieba-Palus, Examination of spray paints on plasters by the use of pyrolysis-gas chromatography/mass spectrometry for forensic purposes, J. Anal. Appl. Pyrolysis 86 (2) (2009) 252–259. [10] C.D.J. Krausher, Characteristics of aerosol paint transfer and dispersal, Can. Soc. Forensic Sci. J. 27 (3) (1994) 125–142. [11] D. Marin, N. Berger, P. Buzzini, G. Massonnet, Transfer, detection and in situ raman analysis of spray paint traces on clothes, in: Proceedings of the 10th European Paint and Glass Group Meeting, Prague, 2004. [12] A. Zeichner, N. Levin, E. Landau, A study of paint coat characteristics produced by spray paints from shaken and nonshaken spray cans, J. Forensic Sci. 37 (2) (1992) 542–555. [13] C. Muehlethaler, G. Massonnet, M. Deviterne, M. Bradley, A. Herrero, I. Diaz de, Lezana, S. Lauper, D. Dubois, J. Geyer-Lippmann, S. Ketterer, S. Milet, M. Bertrand, W. Langer, B. Plage, G. Gorzawski, V. Lamothe, L. Marsh, R. Turunen, Survey on batch-to-batch variation in spray paints: a collaborative study, Forensic Sci. Int. 229 (2013) 80–91. [14] C. Muehlethaler, G. Massonnet, P. Esseiva, The application of chemometrics on infrared and raman spectra as a tool for the forensic analysis of paints, Forensic Sci. Int. 209 (2011) 173–182. [15] E.N.F.S.I. European Paint, Glass Group, Manual of best practice for the forensic examination of paint. Appendix E, Raman Spectrosc. (2009). [16] D. Lambert, C. Muehlethaler, L. Gueissaz, G. Massonnet, Raman analysis of multilayer automotive paints in forensic science: measurement variability and depth profile, in: Proceedings of the 7th International Congress on the Application of Raman Spectroscopy in Art and Archaeology, Ljubljana, 2013.