You are on page 1of 5

Experimental Design for Formulators

SpecialChem | Edward M Petrie - Jun 30, 2010

Value of DOE to the Formulator


A Simple Mixture DOE
Additional Values
Examples of Adhesive Formulations Employing DOE

Everyone is trying to operate faster and more efficiently these days. Adhesive formulators go to great
lengths to offer adhesive formulations that "cure on demand" or provide an "instantaneous bond" for their
customers interested in reducing production time. Yet, in their own shops many formulators often employ
time consuming, inefficient processes such as "trial-and-error" in developing these same formulations.

The design of experiments (DOE) approach to optimizing formulations is a powerful tool for formulators to
find an ideal formulation via statistical analysis of mixtures. This article illustrates how to apply DOE
methods for the purpose of accelerating product development and discovering formulation relationships
that may remain hidden using a standard trial-and-error approach. A simple example is used to illustrate a
DOE mixture problem, and references are provided for those wanting to go deeper into statistical
methodologies.

Value of DOE to the Formulator

The generation of improved and customized product formulations is essential for adhesive companies to
ensure cost effective, high quality, and marketable products. These products must be adapted to the
specific market situation and the needs of the customer. As a result countless formulations may be
required to meet the varied demands.

In addition to product development, the adhesive formulator must be able to periodically cost reduce his
or her formulations to stay competitive. Often an old technique, known as "cherry picking" is used. This is
the process by which an attempt is made to replace one or two higher cost raw materials with less
expensive alternatives by direct substitution in the formulation, leaving other ingredients alone. This is a
somewhat more directed approach, but still a "trial-and-error" process.

The problem is that every formulation is a complex, interactive system, and the improvements that can be
made are often due to the interaction of materials with one another. This means that the optimal
formulation is one in which the levels of some ingredients are increased while others are lowered - an
outcome that cannot be easily arrived at by trial-and-error. If a formulation has 15 to 20 components, and
just two levels of each ingredient are examined, there are millions of possibilities to consider.

Fortunately, a number of statistically based methods are available to assist the formulator in this
optimization process. The primary instrument is known as the "design of experiments" (DOE) process.
This tool allows us to understand the entire universe of possibilities but with only a small subset of testing.
More than one variable at a time can be examined with statistical experimental design, and interactive
effects that are subtle and difficult to detect are revealed. The DOE and trial-and-error methodologies
each have their advantages and disadvantages and these are shown in Table 1.
Trial-and-Error Approach Design of Experiments Approach
Advantages Disadvantages Advantages Disadvantages
 Allows for
 Data captured on
mathematical
only the
optimization of
experiments  Requires highly
performance and
actually run reproducible,
 Can be faster cost
 Highly operator quantitative test
 Tests are  Data captured for
dependent methods
qualitative in an unlimited
 May not provide  Requires training in
nature number of
the best answer statistics and
variations
 Cannot experimental design
 Data can be
simultaneously
repeatedly
optimize cost
searched

Table 1 : Formulating Approaches: Advantages and Disadvantages

A Simple Mixture DOE

A simple mixture approach to formulation development can illustrate the principles of the design of
experiments approach. The example used here is derived from that provided by Anderson and Whitcomb,
which is an excellent fundamental description of a simple three component mixture DOE. 1

Planning a mixture experiment typically involves the steps shown in Table 2. One does not necessarily
have to start at the beginning since a lot of information may already be available that can feed the
process. The statistical analysis approach could be supplemented with a database of established product
formulations and day-to-day manufacturing variations to even further reduce the amount of testing
required.

1. Define the objectives of the experiment (e.g., minimize viscosity)


2. Select the mixture components and any other factors to be studied (other factors may include
process variables)
3. Identify any constraints on the mixture components (e.g., percentage of all components must
equal 100%)
4. Identify the response variable to be measured (e.g., viscosity)
5. Propose an appropriate model for modeling the response data as functions of the mixture
components
6. Select an experimental design that is sufficient not only to fit the proposed model, but which
allows a test of model adequacy as well.

Let us consider an adhesive system that requires a minimum viscosity, and that a blend of three solvents
can be used. If the properties are dependent only on the component ratio of the solvents, the mixture
relationship can be expressed as:

X1 + X2 + X3 = 1.0 (or 100%)

The expression X can be molar, weight or volume fraction or percent concentration of each solvent in the
solvent blend. In this case, there are two independent variables, since the level of one of the ingredients
is set by the fact that the solvent concentrations in the blend must add up to 100%.

Table 3 shows the experimental design in a convenient tabular format that identifies the blends to be
tested by their type or formulation identification. These experiments are then run in order to determine the
response factor (solubility, g/l) resulting from the various solvent blends. The actual order in which the
experiments are run should be randomized to counteract time-related effects such as shelf-life of the
components, temperature in the laboratory, operator experience, etc.
Formulation Blend Component A, Component B, Component C, Resulting Measurement
Identification Description % % % (Solubility, g/l)
1 Pure A 100 0 0 121
2 Pure B 0 100 0 164
3 Pure C 0 0 100 179
4 Binary AB 50 50 0 140
5 Binary BC 0 50 50 180
6 Binary AC 50 0 50 185
7 Equal 33.3 33.3 33.3 199
8 Check-1 66.6 16.7 16.7 175
9 Check-2 16.7 66.6 16.7 186
10 Check-3 16.7 16.7 66.6 201
Table 2 : Steps Involved in Mixture Experiments

For three raw materials, eleven experiments are generally necessary (2n + n where n equals the number
of components). Ten experiments are necessary in this example due to the fact that one degree of
freedom is lost because of the constraint that all solvent components must equal 100%.

Graphical illustrations of the experimental space is desirable. For a three component mixture, this can be
represented by an equilateral triangle. At any point in the triangle the mixture design rule is obeyed. The
experimental region for the example experiment is illustrated in Figure 1. For a four component mixture
another dimension is required for graphical illustration, and a tetrahedron is used. For a larger number of
components, the modeling can be done via computer.

Figure 1: Graphical representation of mixture experiment

There are many mathematical models that can be used for the analysis and representation of mixture
data. Generally, least squares regression analysis is used to develop a polynomial equation that best
describes the resulting data. 2 This regression analysis is generally completed via a statistical computer
program of which many are commercially available. For the example experiment, the response mixture
model is:

Y (solubility) = 122A + 165B + 178C - 6AB + 141AC + 35BC +799ABC

where A, B, and C are the component concentrations.


In this equation the coefficients of the first order terms (A, B, and C) represent the response from the pure
components. For example, solvent C is the most efficient solvent due to its relatively high coefficient
(178). The second order coefficients (AB, AC, and BC) represent interactions. Positive coefficients
suggest synergy and negative coefficients suggest opposition. Third order terms (ABC) represent three
component interactions.

Contours of response, generated by the model fitted to the data, can be plotted (Figure 2) as a visual aid
for interpretation of the results. This can also show results for mixtures that have not yet been tested.
Sometimes rather unexpected results can show up on the contour plots.

Figure 2: Response surface graph.1


(SpecialChem Fig. Ref.: Response graph)

The graphs provide valuable information about the formulation and show expected results from mixtures
that were untested. From Figure 2, for example, it is apparent that an optimum combination of solvents for
the minimum viscosity (or greatest solubility) is 27% of component A, 26% of component B, and 47% of
component C.

Additional Values

Mixture DOE can also be used to reduce costs of formulations. Many successful products have raw
material costs that are often 5 to 15 percent higher than necessary. This is usually because the product
formulations are never cost optimized before going to market. DOE can be used to develop mathematical
optimization of both performance and cost. By including cost data in the experimental process, the
formulators can focus on optimizing the value of the formulation (performance / cost) rather than merely
performance.
One must also realize that one of the factors that goes into the overall price of the adhesive is the
development cost. By reducing development costs via DOE experimentation, the formulator can address
the needs of his or her customer in a faster, less costly, and more efficient manner than if only "trial-and-
error" approaches were used for experimentation.

The design of experiments approach can also be applied to processing variables as well as to formulation
variables. Consider the adhesive bonding process where the primary concern of interest is the strength of
the bond and its variation. Through experimentation, one can predict the bond strength under various
conditions such as cure time, temperature, and pressure.

Examples of Adhesive Formulations Employing DOE

Several recent examples point to the time and cost savings that can be achieved by such processes.
Stepan Company has developed a design of experiments "mapping" approach to help formulators
discover how their products work in specific applications. 3 They used a DOE software program (Design-
Expert) supplied by Stat-Ease Inc. to produce contour plots of different mixtures. Armed with this design
map, adhesives chemists can review the typical properties of complex formulations including open time,
green strength, and substrate bonding profiles.

Another example of the use of DOE involves the cost reduction of a two-component epoxy adhesive
formulation that had been in the marketplace for many years. 4 Four alternative formulations were
identified that offered a raw material cost savings of 14-18% and had equivalent or better performance.
This task took place over only four months and required testing of 90 formulations. One of the highlights
of this program was that two of the original raw materials were found to be unnecessary and were
eliminated from the product.

Experimental design was used to put in order the formation factors of an epoxy adhesive. Factors such as
the nature of the epoxy, the hardener, plasticizer, fillers, and cure promoter were investigated for reactivity
and strength. 5 The work shows that the "one-component at a time" experimental strategy is too time
consuming for a systematic study of each formulation factor.

Segura Velandia, et. al. describes how existing information can be used in developing new formulations. 6
The traditional approach to reuse existing expertise often lacks consistent and systematic procedures. A
technique called "Case Based Reasoning (CBR)" is used to predict polyurethane formulations from
existing data. This approach has shown to have the potential of reducing the time and cost of product
development while at the same time enabling simulation studies. It also serves as an "institutional
memory" to support problem solving.

References

1. Anderson, M.J. and Whitcomb, P.J., "Mixture DOE Uncovers Formulations Quicker", Rubber and Plastics News,
October 21, 2002.
2. Scheffe, H., "Experiments with Mixtures", J. of the Royal Society Series B, Vol. 20, No. 2, 1958, pp. 344-360.
3. Hillshafer, K., "Formulating by Design", Adhesives and Sealants, March 2004.
4. Rudolph, S.E., "Finding the Goldmine in a Product Line", Adhesives and Sealants, October 2005.
5. Grohens, Y., et. al., "Experimental Design as a Route for Improving the Performances of Formulated Epoxy Adhesives",
Polymer Testing, Vol. 16, 1997, pp. 417-427.
6. Segura Velanida, D.M., Heath, R.J., and West, A.A., "Formulating Polyurethanes Using Case Based Reasoning", Plastics,
Rubbers, and Composites, Vol. 36, No. 6, 2007.

You might also like