Figure 2. A typical output of a simulation model shows cold room capacity.

"Scope creep" is very common when the model is demonstrated to different groups. With each demonstration comes a request to add yet another analysis or feature. This results in a time-consuming development effort and introduces difficulties in the operation of the model once it is complete. A complicated model is more susceptible to errors, slower to run, and usually requires more detailed inputs. This means that changes to inputs, either as part of standard maintenance (such as process changes, production mix, and volume changes) or as part of scenario analysis and planning, become more frequent and difficult to make. The result is a model that is less flexible and more difficult to use and may not be an effective planning tool overall. Additionally, a complicated model may be more difficult to verify since troubleshooting model inaccuracies can be challenging.

It is important to keep the model as simple as possible. With input from all key stakeholders, review the process flows and create a simplified version of the process that focuses on the most critical resources and operations. Avoid going to the other extreme of over simplification by getting concurrence of all key stakeholders.

Needless to say, potential errors must be carefully reviewed when making such assumptions. A piece of equipment that is not considered a bottleneck may turn out to be one (the simulation model can actually help in identifying these), or a noncritical operation may be keeping resources away from a critical one. Use care when deciding what should and shouldn't be modeled. In many cases, a small first-pass static model can be a good starting point for the more complex simulation model.

GENERAL VS. PRECISE DATA Similar to simplifying the flow, one needs to consider the amount of detail to be included within each step of the model. Most simulation packages allow a significant amount of detail for each operation. One example is the definition of the production demand for a modeled operation. You can define specific dates for the schedule (for example, "start lot A on Jan-5, lot B on Jan-9") or define demand on an interval-based schedule ("start one lot every 4 days"). Other examples are the size attributes that are assigned to production lots. Each lot can be specified in detail ("lot A is 50,000 vials of product AAA") or have attributes assigned by the model using a distribution based on the production mix ("on average, lots of AAA will be 50,000 vials with a random distribution").

While the use of precise data allows for more detailed planning and analysis based on the real production forecast, it also requires more data processing and consideration of factors that could otherwise be ignored or simplified (such as holidays, shift structures, and rotating schedules). The pros and cons must be considered against the objectives of the model to determine the right approach.

MODEL VERIFICATION As with any other model, once the simulation model is created, it should be put through a rigorous verification process to guarantee its performance and accuracy. One good verification technique is to use historical performance data to test the model. After plugging in previous production plans, the user can then compare model results with actual production performance

Some simulation models are not robust. You must test the model in conditions that exceed the range of normal parameters (for instance, additional or fewer resources and higher or lower production volumes).

Since a simulation model considers variability that can randomly sway outcomes, a successful model should not be expected to reflect real-life performance case-by-case. Rather, it should show an accurate trend or average performance. For example, the cycle time for a specific lot may be different in simulation than on the floor due to built-in variability, but the average cycle time for several lots should be more accurate.

One of the key inputs for such models is not only average labor and equipment processing times for a given product, but the variability of those times. As in real life, the average is rarely, if ever, achieved. A model that accurately accounts for this reality can provide a better picture of the production needs.

MODEL MAINTAINANCE Over time, the modeled operation will change, and a formal process should be put in place to update the model. A valuable technique to make this manageable is to assign a model owner in charge of tracking all desired changes. These changes are reviewed semiannually or annually in the context of future operational strategy, and a short design process is initiated to clearly define the scope of the updated model.

Additionally, process data should be collected on an ongoing basis in order to maintain model integrity. Systems should be put in place to capture actual processing characteristics (averages and standard deviations) of key model steps. These figures should be periodically reviewed, and the model should be updated accordingly.

Simulation is a powerful tool that can offer many benefits if done right and used appropriately. Like most powerful tools, this one also needs to be handled with care. Knowing what the model is expected to achieve and correctly identifying the relevant variables and their behavior is the key to ensuring meaningful results.