Using Near-Infrared Diffuse Reflectance Spectroscopy to Quantitatively Predict of Components of Tobacco

The tobacco industry relies on chemical analysis as it is vital that the processing plant can determine the composition of raw tobacco. This varies depending on where the tobacco was grown and what the climate was like during that time, and gives an understanding into how to maintain product consistency during processing and in the design on few products. Another important factor is the amount of chlorine in the tobacco which can make difficulties in the testing lab; some processors will not provide full compensation to growers if the chlorine content is too high.

The traditional automatic analyzers (e.g. continuous segmented flow analyzers) have been the standard technique for quality control laboratories in the tobacco industry, but these methods require specialized operators and are costly and time-consuming both in analysis and upkeep. Automatic analyzers also necessitate the use of hazardous chemical reagents for colorimetric determination of some tobacco components, which produces harmful waste that needs protective equipment to be disposed of properly.

For example, to determine nicotine content the color is measured which results from the tobacco reacting with sulfanilic acid and cyanogen chloride, which is generated in situ by the reaction of chloramine T (which needs to be prepared fresh every five days) and potassium cyanide (which needs to be prepared fresh every two weeks), both of which are hazardous chemicals.

Near-infrared (NIR) spectroscopy is quick, low cost, offers simple analysis, and can analyze multiple components with one spectrum using chemometric models. This makes it an ideal complementary analytical technique to run automatic analyzers for large-volume quality control analysis of tobacco.

NIR spectroscopy can also analyze a sample without destroying it, allowing for reuse of the sample and a reduction in cost and waste compared to wet chemistry techniques. This is especially useful for geographic regions susceptible to drought, where access to water for sample preparation can be restricted, causing delays in analysis and production costs to skyrocket.

This article will discuss the ability to accurately predict the amount of nicotine, sugar, and chloride in tobacco using the i-Spec® Plus, a portable diffuse reflectance NIR spectrometer with on-board chemometric software from B&W Tek.

Experimental

Instrumentation

Samples of dry, ground tobacco were provided over one growing season by a South African company. This company grows and processes tobacco from different regions within South Africa and along with various curing conditions, resulting in tobacco samples that can have a wide variety of nicotine and sugar contents.

The on-board software of the i-Spec® Plus 2.2 Diffuse Reflectance NIR Spectrometer took NIR measurements (Figure 1), measuring 1100-2200nm with a resolution of 10 nm. All tobacco samples were placed in a petri dish and analyzed on a turntable adapter on top of the i-Spec® Plus so as to take into account any sample heterogeneity. Integration times were 600 µs and averaged 4,000 times per acquisition for a total acquisition time of 2.4 seconds.

i-Spec® Plus Diffuse Reflectance Portable NIR Spectrometer with turntable

Figure 1. i-Spec® Plus Diffuse Reflectance Portable NIR Spectrometer with turntable

Chemometrics

Predictive chemometric models of three main tobacco components: nicotine, sugar, and chloride, were created using BWIQ® software.

In order to formulate a calibration model, flue and air cured varieties of tobacco were used. The methods used for each reference value were segmented flow analysis for nicotine and sugar ad potentiometric titration for chloride. These were provide by the company and are based on the British American Tobacco (BAT)-approved methods for measuring content.

The information for the tobacco samples used to develop the three chemometric models is displayed in Table 1. Each model contains sample that give full representation of typical nicotine, sugar, and chloride concentrations over the different curing methods and also include some samples with out of specification values.

Table 1. Tobacco sample information

Tobacco
component
# of calibration
samples
% content
range
# of validation
samples
Nicotine ~2,000 flue-cured
~700 air-cured
0.091-7.37 250
Sugar 0.28-25.12
Chloride ~1,200 flue-cured
~300 air-cured
0.12-8.29 150

 

Results

Figure 2 shows the raw NIR spectra files acquired on the i-Spec® Plus imported into the BWIQ® software. A Leverage test was used alongside the software to detect any outlier spectra in the calibration samples. These outliers were removed from the data set before the models were developed.

Tobacco NIR raw spectra collected on the i-Spec Plus system

Figure 2. Tobacco NIR raw spectra collected on the i-Spec Plus system

There is a degree of offset in the baseline of the spectra, seen in Figure 2, due to scattering effect in the tobacco samples. In an effort to minimize this offset, a Multiplicative Scatter Correction (MSC) treatment was applied to the calibration sample set in the BWIQ® software. The software was also used to further minimize the baseline and slope difference and to enhance the signal variation in the calibration set by applying a second derivative pre-treatment. The effect of these treatments can be seen in the post processing spectra in Figure 3.

Processed NIR spectra (MSC and second derivative). Wavelength ranges were isolated to 1198-2050 nm for sugar and nicotine and 1800-2050 nm for chlorine.

Figure 3. Processed NIR spectra (MSC and second derivative). Wavelength ranges were isolated to 1198-2050 nm for sugar and nicotine and 1800-2050 nm for chlorine.

There have been earlier studies that use NIR for tobacco processing that have shown that wavelength selection is vital in building predictive models as these wavelengths have to correspond directly to the components being detected. Furthermore, although second derivative pre-processing enhances the variation in the data, it also amplifies background noise present in the spectra.

Using BWIQ® for manual wavelength selection allows the selection of only specific wavelengths, as well as eliminating the spectral region above 2050 nm to remove the pre-treatment enhanced noise. This study modeled nicotine and sugar contents with a wavelength range of 1198-2050 nm. Chloride was modeled with a wavelength range of 1800-2050 nm.

(a) PLS calibration curve for nicotine (b) PLS calibration curve for sugar and (c) PLS calibration curve for chloride. Blue indicates calibration samples, red indicates validation samples.

Figure 4. (a) PLS calibration curve for nicotine (b) PLS calibration curve for sugar and (c) PLS calibration curve for chloride. Blue indicates calibration samples, red indicates validation samples.

All calibration sets had a partial least squares (PLS) regression algorithm applied for quantitive analysis. Figure 4 shows the generated calibration curves for (a) nicotine, (b) sugar and (c) chloride components.

BWIQ software reports several chemometric parameters that are indicative of the performance of a calibration model. The linearity of the calibration curve can be described by Pearson’s R coefficient, with RMSE standing for Root Mean Square Error of the calibration and validation. The chemometric parameters that describe the quality of the created calibration curves for nicotine, sugar, and chloride in tobacco can be seen Table 2.

Table 2. Regression coefficients for nicotine, sugar, and chloride models

Component Number of factors Pearson’s R coefficient RMSEC RMSEV
Nicotine 6 0.95699 0.37890 0.42200
Sugar 8 0.93895 2.14924 2.13391
Chloride 8 0.95635 0.48863 0.51023

 

The quality of predictions using the created calibration curve is set by the validation sets. To better show the difference in the prediction value against the reference value of the validation samples, the residual distribution plots were analyzed. The residual plots of for the nicotine, sugar, and chloride models are seen in Figure 5. If the models have good linear agreement, the samples should lay close to the zero line of the residual plot (red line), and be evenly distributed around it.

Residual plots for (a) nicotine, (b) sugar, and (c) chloride models.

Figure 5. Residual plots for (a) nicotine, (b) sugar, and (c) chloride models.

Conclusion

B&W Tek’s i-Spec® Plus portable NIR spectrometer has shown that the contents of nicotine, sugar, and chloride in ground tobacco can be quantified quickly and accurately using models developed with BWIQ® chemometric software. Once a chemometric model is created for the factors of interest, it can be used to accurately predict the content within a tobacco sample, allowing for rapid, on-line quality control analysis. The ability to rapidly analyze large volumes of tobacco samples using a NIR system removes the need for hazardous chemicals and the waste that is produced by using traditional wet chemistry techniques.

This information has been sourced, reviewed and adapted from materials provided by B&W Tek.

For more information on this source, please visit B&W Tek.

Citations

Please use one of the following formats to cite this article in your essay, paper or report:

  • APA

    B&W Tek. (2019, January 09). Using Near-Infrared Diffuse Reflectance Spectroscopy to Quantitatively Predict of Components of Tobacco. AZoM. Retrieved on May 21, 2019 from https://www.azom.com/article.aspx?ArticleID=15961.

  • MLA

    B&W Tek. "Using Near-Infrared Diffuse Reflectance Spectroscopy to Quantitatively Predict of Components of Tobacco". AZoM. 21 May 2019. <https://www.azom.com/article.aspx?ArticleID=15961>.

  • Chicago

    B&W Tek. "Using Near-Infrared Diffuse Reflectance Spectroscopy to Quantitatively Predict of Components of Tobacco". AZoM. https://www.azom.com/article.aspx?ArticleID=15961. (accessed May 21, 2019).

  • Harvard

    B&W Tek. 2019. Using Near-Infrared Diffuse Reflectance Spectroscopy to Quantitatively Predict of Components of Tobacco. AZoM, viewed 21 May 2019, https://www.azom.com/article.aspx?ArticleID=15961.

Ask A Question

Do you have a question you'd like to ask regarding this article?

Leave your feedback
Submit