Differentiation of used motor oils on the basis of their IR spectra with application of cluster analysis
Introduction
Samples of motor oils and car lubricants that are revealed on the clothes of car accident victims are frequently the subject of investigation for forensic purposes. The identification of an oil sample from an oily stain on clothes occurring as the result of contact with a car can contribute to the apprehension of a perpetrator of an accident. Although the chemical contents of various types of motor oils significantly differ both qualitatively and quantitatively (in terms of the set of additives [1], [2]), the identification of an oil is complex due to the degradation processes it undergoes during vehicle exploitation. Application of IR spectrometry and elemental analysis enables one, however, to differentiate samples of oils of various degree of use [3], [4], [5]. The differences observed in the IR spectra of used oils are very often small and to assess their significance, i.e. to establish whether they originate from real differences or from statistical errors, it is necessary to apply an appropriate statistical method.
Chemometric approaches such as factor analysis and cluster analysis are widely used in molecular spectroscopy [6], [7], [8], [9], [10]. They are also being used with increasing frequency in the field of forensic science [11]. Until now there has been no information in the literature concerning the application of these methods to the differentiation of the chemical compositions of used oils on the basis of their IR spectra. An attempt to solve this problem made by the authors earlier by means of the correlation method did not yield fully satisfactory results [12]. Thus, the aim of the current work was to check the usefulness of the above-mentioned chemometric methods.
Section snippets
Experimental
Two kinds of motor oils were studied:
Elf Sporti Super Oil used in an OPEL ASCONA.
Castrol GTX 3 Oil used in a FORD ESCORT.
Both oils were sampled during the course of usage in the cars. The samples were taken directly from the sumps of the cars at strictly defined intervals of kilometres driven:
0, 2–50, 1000, 1500, 2000, 3000, 4000, and 6000 km when Elf Sporti Super Oil was used, and
0, 2–50, 500, 1000, 1500, 2000, 3000, and 4000 km when Castrol GTX 3 Oil was used.
Infrared (IR) spectra were produced
Mathematical procedure
The procedure applied for processing and interpretation of the measurement data (IR spectra) was as follows:
- •
A spectrum was considered as a vector X:where j is the number of a given spectrum and xi,j represents the value of absorbance measured for this spectrum at wavenumber λi. For each spectrum as many as 299 absorbances distributed regularly along the wavenumber scale within the ranges 600–1300 and 1500–2000 cm−1 were taken into account.
- •
Spectrum Xj was transformed into the
Results
Individual spectra obtained for all samples of both examined oils are shown in Fig. 1.
The results of the factor analysis with reference to both groups of spectra examined are presented in Fig. 2. It can be seen that only three factors are characterised by almost the same total value of variance as the raw vectors of absorbances. Therefore, it was decided to perform a comparative analysis of oils taking account just this number of factors.
The visualisation of particular spectra of both oils in
Discussion
The choice of Elf Sporti Super and Castrol GTX 3 oils for the comparative study described in the current work was not accidental. They represent those kinds of oils that undergo chemical changes during their exploitation either relatively fast (Elf) or very slowly (Castrol). Thus it seemed to be interesting to check the effectiveness of cluster analysis in a comparison of series of IR spectra from these two extreme cases.
A visual analysis of the appropriate spectra obtained for the oils of
Conclusions
One can conclude that — unlike in other methods of analysis of the similarity of objects (e.g. the correlation method [12]) — in the chemometric approach applied in this work, the IR spectrum is treated as a set of analytical data without its physical and chemical interpretation. Moreover, cluster analysis does not allow us to perform a statistical evaluation of the revealed effects. For this reason it is used mainly in the exploratory stage of a study, when no a priori hypothesis is available
References (16)
- et al.
Forens. Sci. Int.
(2000) Forens. Sci. Int.
(1998)- et al.
J. Mol. Struct.
(1999) - et al.
Probl. Forens. Sci.
(1994) - et al.
J. Forens. Sci.
(1994) - et al.
Appl. Spectrosc. Rev.
(1996) - et al.
Appl. Spectrosc. Rev.
(1996) - et al.
Appl. Spectrosc. Rev.
(1997)