Chaughule, Thorat - Statistical Analysis/Design of Experiments
complex and many factors can influence the desired response then statistical tool like
Plackett-Burman design can be useful. In food dehydration process, Plackett-Burman
design can be used for screening of most influential factors on the response. If the
process is not optimized by first order models various second order modeling designs
such as Doehlert design, Taguchi design, Box Behnken design, response surface designs
can be employed. The choice of the design can be made by characteristics of the desired
response. Taguchi design is useful for minimizing noise factors i.e. the factors which are
affecting the extent of response. For example in case of hot air drying of banana slices, if
the response is optimized for certain values of air temperature and velocity. The fluctua-
tion in relative humidity is the cause of variation in experimental settings then it is
called as noise. For such problems Taguchi design is a useful tool. Response surface de-
sign is the most widely and popular design for optimization of multivariate systems.
Usually processing processes affected by 3-4 factors are considered. Like all other de-
sign RSM suggests set of experiments for different levels of factors. Response surface
plots can be obtained from the experimental data by fitting the second order polynomial
equation. RSM plots show surface interactions amongst the factors for obtaining the de-
sired response. However, RSM suffers from some drawbacks such as large variations in
the factor levels can be misleading and can result in to error or bias. Sometimes critical
factors may not be defined properly and sometimes over reliance on computer. Unlike
RSM if the extreme levels for factors are not desirable then Box Behnken design can be
applied. In EVOP two or more factors can be varied at a time, hence can be used in the
complex systems influenced by many factors. In some cases Doehlert design may find
application as it takes minimum number of experiments as compared to other second
order model design. In case if one design does not satisfy the optimum point, different
second order polynomial can be screened for desired optimum response.
There is no general rule for selecting a statistical design for drying of foods, vegeta-
bles and fruits. The choice of design in many cases is decided on response characteristics
and feasibility of experimentation. Various DOE software are available commercially
which can be employed for the above mentioned designs.