CORESTA Congress, Paris, 2006, SS 13

A mixture experimental design with six variables and the effects on smoke components:1 - Presentation of this experimental design and of some initial results on physical characteristics of cigarettes and blends

VIDAL B.; FIGUERES G.; BIESSE J.P.; BREGEON B.; LOUVET F.; MUZELLEC L.
Altadis Research Centre, Fleury-les-Aubrais, France

The objective of this paper is to examine the effects of six tobacco components on the chemical and physical characteristics of blends and cigarettes and particularly on the mainstream smoke yields. First of all, we will develop the philosophy of this design, then we will discuss about the interpretation of some initial findings and their limits. In a second paper, we compare the effects of the six components on Hoffmann analytes. Often, we can notice different experimental designs on the blend components with three factors, but to obtain a more realistic view of the whole blend, we have carried out an experimental mixture screening design with a regular simplex with six factors on one US blend. These six factors are the content of: Flue-cured block (from 50.7% to 13.5%), Burley block (from 56.9 to 19.7%), Oriental block (from 42.2 to 5.0%), the Recon (from 49.8% to 12.6%), Stems (from 41.2% to 4%) and Expanded Tobacco (from 45.2% to 8%). The experimental design obtained is built with reference blend (centroïd) and consists of 19 trials, and two added trials for validation. We have added the six pseudo components with the aim to test the additive effects of the different variables' responses. We can classify the different determinations in chemical variables (n = 38), physical characteristics of cigarettes (n = 17) and smoke variables (n = 43). On some examples, we will show that the statistical analysis allows, first of all to compare the relative effects of the six factors on each response variable by the graph trace, then, to calculate a polynomial model of prediction and its quality and ability to optimize the responses and finally to get correlations between factors and variables and to try to identify a structure on multidimensional analysis mapping (P.C.A and cluster analysis).