DEVELOPMENT OF AN EXPERT-SYSTEM FOR SELECTION OF EXPERIMENTAL-DESIGNS

An expert system has been developed to assist chemists in the selection of experimental designs for research projects. The system (named DXPERT) ranks thirteen types of experimental designs according to their suitability for projects presented by users in an interactive session. Design categories included are factorial, response surface, sequential simplex optimization, simplex mixture, and statistical testing. A desirability index is assigned to each design alternative according to project characteristics (attributes). Characteristics are interpreted based on expert knowledge built into the system. DXPERT uses mathematical concepts to mimic features of human intuition and decision making. Expert knowledge is represented by relevance factors (a number between minus one and plus one) in a multiple-alternative multiple-attribute table. Relevance factors are interpreted as fuzzy values that represent the degree to which a design belongs to the set of suitable designs. The formula for calculating design desirabilities is based on fuzzy mathematics. For efficiency purposes, the order of the questions presented to the user is driven by a maximum potential information gain algorithm. Design desirability indexes were found to be useful to researchers for the elimination of unsuitable designs and concentration of further efforts in the most applicable designs. A validation test was conducted with the participation of four other experts in the field.