Review: Characterisation and Classification of Automotive Clear Coats with Raman Spectroscopy and Chemometrics Forensic Purposes

Emily C. Lennert

Category: Chemistry

Keywords: automotive, vehicle, car, paint, clear coat, spectroscopy, Raman

Article to be reviewed:

1. Maric, M.; Bronswijk, W. V.; Pitts, K.; Lewis, S. W. “Characterisation and classification of automotive clear coats with Raman spectroscopy and chemometrics for forensic purposes.” Journal of Raman Spectroscopy. 2016, 47 (8), 948–955.

Additional references:
2. What is Raman Spectroscopy? (Accessed October 31, 2016).

Disclaimer: The opinions expressed in this review are an interpretation of the research presented in the article. These opinions are those of the summation author and do not necessarily represent the position of the University of Central Florida or of the authors of the original article.


Automotive paint, in the form of chips or transferred paint/smears, is a form of physical evidence that is often recovered from automotive related incidents, such as crashes or hit-and-runs. In some cases, automotive paint may be the only actionable evidence recovered from the scene. Although no universally accepted method exists for paint analysis, the American Society for Testing and Materials (ASTM) and the Scientific Working Group on Materials Analysis (SWGMAT) have put forth guidelines for paint analysis. The guidelines set forth by ASTM and SWGMAT outline techniques for several instruments, including infrared spectroscopy, elemental analysis techniques, Raman spectroscopy, and more. A known sample source, such as a suspect’s vehicle, can be compared to the unknown paint sample recovered from the scene. In instances where a known sample is not available, an examiner may compare the unknown paint sample to paint samples maintained in the Royal Canadian Mounted Police’s database. Comparison alone relies on the examiner’s ability to interpret and compare the analytical results, creating opportunity for error. To mitigate this error, the authors of this study utilize chemometrics, the application of statistics to analytical results, after examining samples by Fourier Transform (FT)-Raman spectroscopy. Chemometrics can be used to create groups, or classes, based on statistical analysis of analytical data. Unknown samples may then be assigned to classes based on analytical data.

The authors of this study employ FT-Raman spectroscopy to examine automotive clear coats. Raman is a spectroscopic technique that is complementary to infrared (IR) spectroscopy; in Raman, the information obtained based on molecular vibrations can be used for identification and quantification of a sample.2 Clear coats were chosen due to a lack of research in the area; the authors state that, of the previous research that has examined automotive paints by Raman spectroscopy, most of the previous research has focused either on the pigments in the base coat or the fillers and extenders found in the primer. The authors state that clear coats were chosen because transfer evidence most frequently involves transfer of the clear coat, and clear coats can be examined via handheld instrumentation in situ, i.e. directly on a vehicle.

Samples were obtained from a sunroof fitting company; paint samples were removed from the roof panels that were removed by the company during sunroof installation. The vehicle make, model, year, and vehicle identification number (VIN) was recorded for each sample. A total of 139 individual vehicle samples were obtained from new cars; samples included 17 different manufacturers (makes) and 45 different models. Clear coat samples were removed from the paint samples by cutting small shavings from the surface of the sample. Samples were analyzed by FT-Raman spectroscopy. A total of five spectra per sample were collected. After sample analysis by FT- Raman spectroscopy, the data was processed to prepare for chemometrics.

The first chemometric technique employed by the authors was principal component analysis (PCA). PCA is a statistical technique that identifies sources of variation within the data, those sources of variation are known as principal components (PCs). Based on the variation within the data, i.e. PCs, groupings are create and can be visualized in 3-dimensional PC score plots, as seen in Figure 1 within the study. Plotting of the first three PCs obtained by PCA on the FT-Raman spectroscopy data resulted in 19 groupings. Groupings were then correlated to common traits within the groups, such as country of origin, make, model, and year. A summary of the traits of vehicles within each group can be observed in Table 2 within the study. The authors then compared these results to their earlier work on clear coat analysis performed on another instrument, attenuated total reflectance-infrared spectroscopy (ATR-IR). The sample set was largely the same between the ATR-IR and FT-Raman spectroscopy experiments. The authors state that ATR-IR data resulted in only 8 groupings after PCA, concluding that FT-Raman spectroscopy is a more discriminating method for clear coat analysis.

Linear discriminant analysis (LDA) was used to verify and test the classification system created by the groupings derived from the first three PCs. Based on the three PC score classification model, 96.9% of data was correctly classified when verified by LDA. Only three samples, and their replicates, were misclassified. The authors note that the samples that were misclassified were from single vehicles, thus, the initial classification within the model were not well defined. The authors conclude that the model based on PC1-PC3 cannot assign vehicles to some classes reliably. However, as the number of samples increases and more samples for these misclassified vehicles are entered into the dataset, the authors theorize that the unreliable classifications will increase in specificity and will become more reliable. A test set of the data, representing unknown samples, was also examined by LDA to test the predictive capabilities of the model. Based on the test set, 97.6% of the data was classified correctly, with one sample being misclassified. When compared to LDA from the previous ATR-IR work, the authors note that ATR-IR showed slightly higher classification accuracy. The authors state that the higher accuracy in the ATR-IR model may be due to the smaller number of classes, only 8 groupings compared to 19 by Raman. Although some samples were misclassified during LDA, the authors note that, as a whole, the Raman model displays highly discriminating capabilities.

Scientific Highlights:

  • Raman spectroscopy is one of the instrumental methods recommended for paint analysis in the guidelines set forth by ASTM and SWGMAT.
  • Clear coats are most often transferred in instances where automotive paint is recovered as transferred evidence.
  • Clear coats analyzed by Raman spectroscopy were classified into 19 groups based on PCA. Compared to previous work conducted by the authors on a similar sample set, Raman spectroscopy is a much more discriminating method than ATR-IR.
  • LDA based on PCA groupings confirmed distinct differentiation between groupings.
  • 96.9% of the validation set correctly classified by LDA. 97.6% of the test set, representing theoreticalunknown samples, correctly classified when tested by LDA.

Relevance: In developing classification models, analysts must seek methods that provide high levels of discrimination as well as high rates of classification accuracy in order to be useful in daily casework.

Potential conclusions:

  • Raman spectroscopy can be used to develop a highly discriminating classification model for automotive clear coat analysis.
  • The Raman model provided relatively high classification accuracy, as seen by the LDA results. As the sample set expands, classification accuracy of the Raman model will likely improve.
  • This study only examined clear coats; other components of automotive paint may be used to create additional classification models, allowing further discrimination.