Statistical design and analysis of a test programme to assess the volatility characteristics of ethanol/gasoline blends

摘要

Designing a measurement programme to assess the volatility characteristics of ethanol/gasoline blends posed challenges as the seven base fuel properties of interest were highly constrained and difficult to manipulate independently of one another. The target base fuel matrix was generated by augmenting a 49-fuel fraction of a 7⁵ factorial with 11 additional fuels chosen using D-optimality. Two of the five factors were treated as pseudo-factors and used to generate the levels of two pairs of mutually constrained properties. The 60 base fuels were then blended and subsequently splash blended with 5%, 10%, 15%, 20% & 25% ethanol. The test order for the resulting 360 fuels was structured and randomized to reduce the risk of extraneous sources of variation contaminating the regression models subsequently fitted to the data. Cross-concentration models had to be fitted by generalized least squares techniques as the measured values of the principal dependent variable were structurally correlated.