OCR Text |
Show parameters and their resulting effects. The factors of interest for this research are selected based on the burner design parameters and suspected mechanistic impact on N O x and C O emissions. BACKGROUND A set of experiments was conducted using a two level, factorial design methodology, which determines the effect of multiple parameters on a system (hence the descriptor "multivariate"). This methodology is described in detail in Box, Hunter, and Hunter (1978) and is the basis for a commercial software program, DesignExpert, that was used to design and analyze the factorial experiments. The basic approach is to test various factors (e.g., fuel jet velocity) at two levels (high and low) to determine the effect on a response (NOx). If the factor has an effect on the response, it's response will be distinct and separate from a normal distribution due to random error. Example data are plotted as a normal distribution and on normal probability axes in Figure lto illustrate chance occurrences and main effects. According to the central limit theorem, the data in Figure 1 which fall under the normal distribution (top plot) or along the line (bottom plot) are most likely due to random occurrences; the data which fall outside of this distribution are believed to have an effect which is not due to chance. The advantages of using the two level, factorial design over the typical, one-factor-at-a-time methodology are reduced number of tests and identification of interactions between factors. Normal Distribution Normal Probability CO .o 2 CL N Main Effects Deviaton From Mean Figure 1: Probability Plots Showing Effects |