Information Technology Reference
In-Depth Information
and manufacturing pharmaceutical processes (Chi et al., 2009).
SVM was used as the training algorithm of the developed model
and GA was applied for optimization purposes to obtain target
input settings that would yield optimum system performance.
The process under investigation was the crystallization of an
active pharmaceutical ingredient. The key quality attributes of
an API are identity (physical form), purity (the desired compound
with minimal impurities), and crystal size, as measured by laser
light scattering techniques. The crystallization process contained
12 input variables and 3 output variables, as water amount (x
1
),
water addition time (x
2
), stir rate (x
3
), addition temperature (x
4
), stir
time (x
5
), water temperature for second water addition (x
6
), batch
addition temperature (x
7
), duration of second water addition (x
8
),
cooling duration (x
9
); solvent % (y
1
), XRPD (y
2
), and d90 (y
3
). The
goal was to optimize processing parameters in order to minimize Y
1
and Y
2
, and maximize Y
3
outputs. The overall desirability function
was used as the objective function in the GA. Selection of
chromosomes was random, and new generations were created
both by combination and mutation of chromosomes. The average
overall desirability converged to a fi nal optimum value after about
30 generations. Optimal settings, such as the values of input
parameters obtained, using GA optimization, were in close
agreement to results of Design of Experiments (DoE) optimization.
GAs were further employed to specify a possible and feasible
subset of input settings around the desirable output targets. By
this method, acceptable variation of process parameters was
defi ned. It was also shown that the y
3
output was the most likely to
fall out of the specifi cation limits.
Example 2
Intelligent software was used to predict the relationships between
tablet formulations, roll compaction process parameters, and the
roll compacted ribbon, from which granules for tablet manufacture
Search WWH ::
Custom Search