What is the best way of looking at the simulation warm up period - the period which needs to be left out when looking at the results - i.e.
of the warm-up period in general and in a particular plan simulation model?
two points: normally you don't want to have the time, the model needs to fill with MUs in the statistics time, so you can wait, until the simulation delivers stable output and set this time in the EventController in the field statistics.
If you use random values, plot some important statistical values e.g. statFailPortion. If the graph becomes flat, the warmup period is over.
freelance simulation specialist
Sorry, Steffen, just to summarise,
1.can you please exlain the 1st point in more detail.
2. there is a way of monitoring the simulation via a graph and once that is running flat, the warm up period is over? So, what is the initial simulation period I need to set? 1 day, 1 month or 1 year?
1. check the time in the simulation, when all drains are starting to delete parts in parallel
(otherwise you get e.g. wrong results in the daily output if you have some days without parts at the start of the experiment); at the time you insert at Statistics in the EventController Plant Simulation resets the statistics (start from zero)
2. the average statistical values should be stable after warm-up, if you have random/ stochastic values, then it need some hundred "random events" to get a stable average, if the time between the events is some days (e.g. availability 99.5% MTTR 2 hours --> mean MTBF 400 hours) it may take some years to get a sufficiently large number of events for a good statistical value
freelance simulation specialist
If you need help on a specific case, just send a message!
Determining the warm-up period has a lot of practical aspects. I have attached a basic model that shows how you can use the ExperimentManager in your favour. It also illustrates some of the points made by Steffen.
As you can see there is no output at all in the first few scenarios. Indeed, not a single MU has reached the Drain by then. So this way you can determine the absolute minimum for the warm-up period.
Regarding stability, you have to look at the change in the statistical output (see the report form the ExperimentManager). Oftentimes you will at least use a graphical procedure: just look when the statistical value of interest flattens out. For a more precise statistical assessment you would probably want to use Welch's method, though this is from an age when computer capacity was scarce and they wanted to scrape every bit from the warm-up period they could. On this page there are some useful spreadsheets: http://eu.wiley.com//legacy/wileychi/robinson/students.html
Some other points: 1) if you have some cyclical behaviour in your model, such as a Shiftcalendar, the warm-up period should be in multiples of such a cycle (e.g. 1 or more day/weeks when you use a ShiftCalendar). 2) it is often assumed that steady-state occurs rather quickly, but, as Steffen's example shows, this may take a very long time, e.g. several years. In my experience, factories seldomly remain the same for more that 3 years. So, instability of the statistical output may then simply be an inherent characteristic of the system that you are studying. 3) Be clear in your reporting about how and why you have chosen certain values for your warm-up period.
By the way, this is also a book that we recommend (Simulation: The Practice of Model Development and Use), the latest edition is available from Palgrave. It is very accessable.
Ok, hope this helps!
sorry to get back to you so late. Many thanks for your input and reply.
I cannot open the attached model with my Plant Sim Softwave Version 11.0.3. Would it be possible for you to save it so I can access the model, please?