Semiannual planned shutdown for a facility of this type was 5 weeks, so 21 weeks were available for processing every 6 months.
The time phasing of various facility configurations and staffing levels was evaluated based on that.
Once a baseline scenario for a volume level was run successfully without any variability, probabilistic inputs for most parameters
were created from other similar facilities or equipment vendors. Monte carlo simulation runs incorporating these variability
inputs were used to establish results with statistical confidence intervals. The lyophilization cycle itself was an exception;
time was locked in with one-hour deviation at most. Statistical distributions for deviations were assumed in formulation process
times, which included filling and capping line speeds where people are involved. Figure 2 shows two examples of variability
in filling and capping line speeds for different products. It was also assumed, based on experience, that changeover and manual
cleanings would probably have the most deviation. Unplanned downtime based on time to failure and time to repair was also
incorporated on all equipment, and probabilities of absenteeism were estimated for staffing.
Differences of opinion were voiced among the operations, engineering, R&D, and operations excellence members of the team as
to what filling and capping rates should be used for the design. The mean line rates were key drivers in the volume that could
be expected from the facility. For example, if the manufacturer's stated maximum line rate is 400 vpm, what mean rate should
be used for planning the facility design? The model results became a credible way to judge the reasonableness of the assumptions
about the planned line rates.
Before running the model, engineering work had been done on basic configurations and designs using timelines in spreadsheets.
These did not include the process variability, staffing, or decontamination and cleaning implications of the operation, but
they did rule out certain configurations as unsatisfactory. When model runs were begun, it became obvious that some other
alternatives would not work when all the asynchronous events in the process were taken into account.
Over a two-month period, hundreds of simulations were run to test results of various configurations. Table 1 depicts some
of the configurations that were evaluated in more detail for both the cart and fixed conveyor configurations, with the key
Table 1. Alternate scenarios were evaluated for various equipment configurations, volume levels, and schedules.
Each of these configurations had sets of monte carlo simulations run for them to determine best and worst case results. An
example summary for one of the scenarios is shown in Table 2. It shows summarized results for cart scenario 5 shown in Table
1 with types of volume, wait times, and utilization metrics possible.
Table 2. Each scenario was evaluated based on the resulting performance metrics for the volume and mix.