(303n) Property Based Experimental Design

Product synthesis and design is a complex process of capturing customer's needs and wants in a product that can be produced at an acceptable price point. Often, researchers and manufacturers design products based on core capabilities, adding multiple options and variations in an effort to hit on the attribute the customer truly desires. This design process is as time consuming as it is costly. The main challenge in achieving success is the lack of a defined method, not the inability to identify the desired attribute, or even to identify the important properties of the attribute. As demonstrated from small research groups to large corporations, the product is almost always developed, but usually at a cost above the allocated budget.

It has been shown that the use of Design of Experiments (DOE) methodology efficiently determines the important properties of an attribute and their respective optimum set points using linear and non-linear models in response surface plots. While effective in reducing the number of experiments needed to reach a solution, the method suffers from combinatorial problems when applied to higher order molecular systems. Here, the use of property clustering techniques can improve on response surface plot design by circumventing the combinatorial problems and working exclusively in the property factor space.

Property clustering utilizes a reverse problem formulation in which a system is visualized on a ternary diagram with each axis representing a property. Algebraic and optimization techniques can extend the application range to include more properties, but a limit of 3 properties is used here so that graphical techniques can be employed. Using developed clustering rules, the attribute's properties can be converted into a discrete target, or continuous target region on the ternary diagram. An infinite number of potential candidate molecular species are also represented in the property factor space in similar manner. Solution to the problem is found by using linear mixing rules between the candidate species and the target.

In this research a systematic framework is presented that transforms a desirable consumer attribute into a product mixture with defined, measurable properties. Tools in Design of Experiments are used to determine the important properties of the attribute. To reduce the required experimentation from combinatorial problems, the properties are placed on each axis of a ternary diagram. The desired attribute's properties are converted to a property cluster along with a list of candidate compounds. A solution is found using linear mixing rules. If a solution does not exist with the existing candidates, a second property clustering diagram may be constructed to design a molecule that will. Here the recently developed group contribution method is used to build a molecule using functional groups. The resulting molecule is then placed back into the original property clustering diagram and solved in the aforementioned manner. The solution can be verified to confirm the presence of an optimum using a response surface experimental design.

This contribution will use a case study to highlight the principles of the methodology.