Risk and Variability Analysis Jan 2006
Risk and Variability Analysis Jan 2006
Risk and Variability Analysis Jan 2006
probability distributions and select the BULK PHARMACEUTICAL EXAMPLE involves the chlorination of quinaldine.
output (decision) variables whose values Procedure P-3 (in R-102) involves the
are recorded during the simulation. For Base Case Analysis formation of the product through the
each simulation trial (scenario) Crystal Ball The process analyzed in this example deals condensation of chloro-quinaldine and
generates random values for the uncertain with the manufacture of an active hydroquinone. A side reaction leads to the
input variables selected in frequency compound for skin care applications. The formation of an impurity. The product and
dictated by their probability distributions process has been developed at pilot plant the impurity formed in P-3 precipitate out
using the Monte Carlo method. All input level and it is ready to be moved to large- of solution. The second section of the
variables are perturbed simultaneously and scale manufacturing. Based on input from process deals with product “Isolation and
their interactions are captured through the the marketing department, the objective is Purification”. The precipitate of the
model as fluctuations of the output. Crystal to produce at least 36,000 kg of active product and the impurity formed in
Ball also calculates the uncertainty involved ingredient per year at a cost of no more procedure P-3 is removed in procedure
in the outputs in terms of their statistical than $ 250/kg. P-4 using a Nutsche filter. The product is
properties, mean, median, mode, variance, The entire process model, as it has subsequently converted into a soluble
standard deviation, and frequency been developed in SuperPro Designer, is form in procedure P-5 while the impurity
distribution. shown in Figure 1. The icons in Figure 1 remains in solid form and is removed in
In this study, SuperPro Designer was represent unit procedures (processing procedure P-6. The third section is the
combined with Crystal Ball to perform a steps) and not unique equipment. “Final Purification” section. The product
Monte Carlo simulation of a bulk Multiple unit procedures may utilize the precipitates in procedure P-7 and the
pharmaceutical process. The integration of same equipment at different times. Each precipitate is recovered using a Nutsche
the two tools was made possible by taking unit procedure contains a set of filter (procedure P-8, NFD-101). The
advantage of the Component Object operations that are performed product is then dissolved in isopropanol
Module (COM) technology built in sequentially in the equipment. The (in P-9) and charcoal is added to remove
SuperPro Designer and Crystal Ball’s following equipment items are available certain impurities. The charcoal used for
inherent integration with Excel. The for the large scale manufacturing of this the treatment is removed by Nutche
probability distributions of the uncertain compound: Filtration in P-10. Finally the last section is
input variables were defined in Crystal Ball. – Three 1,000 gal (ca. 3,800 m3) the “Crystallization and Drying” section.
Macros were used to link the uncertain reactors (R-101, R-102, R-103) The product solution is concentrated in
parameters with their corresponding input – Two 4 m2 Nutsche filter (NFD-101, P-11 (isopropanol is vaporized) and the
variables in the SuperPro model. For each NFD-102) product is crystallized (in the same vessel,
set of values of input variables, SuperPro – A tray dryer with a capacity of 1 kg/h R-103). The crystallized product is
performed material and energy balances, removed solvent (TDR-101) recovered using a Nutche filter (P-12)
scheduling, and cost analysis calculations. The process is divided into four sections, and dried using a tray drier (P-13 /
The calculated outputs of SuperPro were boxed with dashed lines on the flowsheet. TDR-101). A more detailed description of
transferred back to Crystal Ball using The fist section is the “Product Synthesis” the process can be found in the
additional macros. section. Procedure P-1 (in R-101) literature (7).
6) Differences in skills of operators that illustrate the impact of the input CONCLUSIONS
affect setup and operation of parameters on the variance (with
equipment. respect to the base case) of the final Deterministic process simulators
7) Availability of operators. process output, when these parameters facilitate modeling of complex systems
Triangular probability distributions were are perturbed simultaneously. This in the chemical and pharmaceutical
assumed for the two main reaction allows us to identify which process industries and provide reliable
operations and the filtration steps that parameters have the greatest correlations between input and output
precede P-9. Even though variability contribution to the variance of the process variables. Monte Carlo
distributions were assigned to specific process and thus focus the effort for simulation tools estimate uncertainty of
operations, it may be deemed more process improvement on them. The output variables from the uncertainty of
accurate to assume that they account sensitivity analysis
for the composite variability of their for the Unit
procedures. If this type of analysis is Production Cost
done for an existing facility, historical and Annual
data should be used to derive the Number of
probability distributions. Crystal Ball has Batches is
the capability to fit experimental data. demonstrated in
The two decision variables Figures 6 and 7
considered in this study are the number respectively. The
of batches that can be processed per duration of the
year and the unit production cost. condensation
These are key performance indicators reaction has the
important for production planning and greatest impact on
project economics. The output the number of
variables, of the combined SuperPro batches and
Designer – Crystal Ball simulation, are consequently the
quantified in terms of their mean, annual throughput.
median, mode, variance and standard If the management
deviation. These results are shown in of the company is
Figures 4 and 5 for the Unit Production seriously Figure 6 – Contribution of uncertain parameters to the variance of the
Cost and the Number of Batches, committed to the unit production cost
respectively. Based on our assumptions annual production
for the variation of the input variables target, it would be
we note that average values (mean / wise to allocate
median / mode) calculated for the R&D resources to
decision variables satisfy the objective. the optimization of
The certainty analysis reveals that we the condensation
can meet the unit production cost goal reaction.
with a certainty of 93% (dark grey area In addition we
of Figure 4). However, the certainty of can see that the
meeting our production volume goal purchasing price
(of 36,000 kg or 147 batches) is only of quinaldine has
83% (dark grey area of Figure 5). Such the greatest impact
findings constitute a quantification of on the
the risk associated with a process and manufacturing cost
can assist the management of a of the final
company in making decisions on product. Focusing
whether to proceed or not with a the market
project idea. research on lower
The dynamic sensitivity charts cost suppliers for
provide useful insight for understanding quinaldine would Figure 7 – Contribution of uncertain parameters to the variance of the
the variation of the process. They be advisable. annual number of batches