Date of Award

Spring 1966

Document Type


Degree Name

Doctor of Engineering Science in Chemical Engineering


Chemical Engineering

First Advisor

L. Bryce Anderson

Second Advisor

John E. McCormick

Third Advisor

Henry Zatzkis

Fourth Advisor

Michael Frederick

Fifth Advisor

C. L. Mantell


A systematic method for selecting from among alternative routes for chemical synthesis is developed and presented. Consideration is given not only to making the best decision, but of reaching that decision with a minimum of experiments.

A method is developed whereby the calculated differences in process cost due to variations in the levels of such responses as chemical yield, usage and recovery are displayed graphically. Such a representation by indicating the potential worth of any experiment focuses attention on those parts of the synthesis which have an economic bearing on the ultimate process cost.

The probability of achieving a particular cost is considered next. Cost functions for each of the proposed routes are developed which relate the effects of the process responses to the process costs. The process responses are in turn related to the settings of the process variables through regression equations. A second order experimental design is generated in the important variables. The data from these design points serve as the basis for the regression equation. The responses to the indicated experiments are estimated listing both a best guess and the range around the best guess. As an experiment is run, the experimentally obtained value replaces the estimated value. The experimentally obtained result and the remaining estimated results are treated identically. The variance about the experimentally obtained value is calculated from experimental error. The variance about the estimated value is calculated from the range about the guessed response. The variance for each of the design points is used to weight the contribution of that point in the analysis.

The results of the analysis are probability distributions for the costs of each process. It is only when the overlap between the probability-cost distributions for competing processes is satisfactorily small that discrimination between routes is achieved. Only by running experiments and replacing estimated results with actual results can the variance about the calculated optimum value be reduced with a concurrent improvement in process discrimination. However, experiments are run only if improved discrimination is required. This feature reduces the total number of experiments required to select preferred routes.

An actual problem is presented in which it is shown how these techniques lead to a process selection on the basis of very few experimental runs.