Model validation
Having created a model, it is necessary to assess its performance; is it producing results that are reasonable? Some of this verification can be done during development; watching the model run at low speed, to observe the behaviour of resources and workpieces, but there remain a number of useful tests that can be performed. Where the model represents an existing system a comparison with the present-day processes is an obvious choice. The model should be able to demonstrate results that are similar to those of the real system, when given the same inputs.
Other methods of validation can also be applied. Assumptions within the model can be subjected to sensitivity analysis; if work-in-progress is always measured at noon, for example, it might be wise to perform a model run where this is measured at some other time of day, to see if the simplification has affected the results.
Static simulation is done by hand, simply using a pen and paper and examining the best possible throughput of in constrained areas. This allows you to estimate the utilisation (and hence, maximum output) of resources, and these figures can be compared with those demonstrated by the model. If a machine is available for eighty hours a week and its operations on a part total ten hours, we would expect to the system’s output to be no more than eight units per week. Thus, even some very simple calculations can serve to provide a reality check. We studied machine utilisation closely, to see if the simulation had the same bottlenecks that production staff had experienced, and this was found to be the case.
We considered the verification process to be complete when performance was normally within 5% of that seen in the focused factory itself. To continue to pursue greater accuracy at this point is to chase a diminishing return, and would have demanded a much more complex model.
Having a valid model was only half the battle, however! This had produced a simulation of the present-day focused factory. We had yet to explore how its efficiency might be improved through alternative methods of logistic control...
