Information Technology Reference
In-Depth Information
can be produced (Mansa et al., 2008). The software exploits the
strengths of ANNs, GAs, and fuzzy logic to predict multivariate
relationships from experimental data. The roll-compaction process
was fi rst modeled to identify relationships between input-output
properties. Input parameters were formulation composition;
excipient particle size distribution; true, poured and tapped density;
Carr's index and Hausner's ratio; compressibility, tensile strength,
effective angle of friction and fl ow function; and roll gap and roll
speed during compaction. Output parameters were ribbon density
and porosity, maximum pressure, and nip angle of roll compaction.
Artifi cial intelligence software, based on neuro-fuzzy logic, was
used to establish input-output relationships. Once the relationships
were appointed, another approach, based on ANNs, GAs, and fuzzy
logic, was used to develop numerically predictive models and
subsequently optimize these models. Neuro-fuzzy rules describe
verbally the relationships between the powder properties, process
conditions, and fi nal outputs, as well as intermediate outputs.
These rules, together with experimental data, were used to develop
and optimize quantitative models. An additional set of experiments
was performed to test and validate developed models. The idea of
an optimization module, based on GAs, is to provide
recommendations of processing conditions on the basis of input
powder properties and desired ribbon properties. For example, the
software calculated that in order to produce a ribbon from dicalcium
phosphate anhydrous with a porosity of 0.35, process conditions
of 1.73 rpm (roll speed) and 0.74 mm (roll gap) were required. The
fi ndings were experimentally confi rmed.
Example 3
GAs were used to optimize non-degradable hydrogel structures as
passive drug delivery systems (Casault et al., 2007). It is diffi cult
to achieve constant drug release from passive platforms, that is,
Search WWH ::
Custom Search