I have this question relating a bit to theory. I know that plant simulation is designed to mimic the working system, but I was wondering that generally when we use it to model a system, how can we verify and validate the model we made?
Any suggestion is much appreciated.
I am also quite interested about this topic. If you are looking for theoritical insights, the paper below can be helpful. The whole paper is very interesting. You can take a look at pages 12 for methods of verification and validation. Hope this helps.
Being not such an experienced user myself, the insights from the experts from this forum about this topic is interesting and something to look forward to.
I guess also here there are a hughe number of theoretical (academic) and practical approaches.
Lets talk about a practical approach.
If you want that somebody pays for your models, then validation is part of your daily work.
Most of the customer accept your model for its behavior, not only for the results. This means, the model should show the same (acceptable accurate) behavior for defined (requested, desired...) system states and the changes between. So you need first learn a lot about the real system (without understanding the real / planned system, you don't need to start modeling). You should be able to predict (observe, ask somebody for) the behavior of the real system in all situations, you will have in the simulation:
- regular processing
- break downs
- shift change
- ramp up, ramp down....
If you create the model and include all the states, you need to check the behavior of the model in each step against the behavior of the real system. Very helpful in this step is to use animation, so you can observe the behavior (e.g. stock development, material movements, worker movements...) and compare it with the predicted behavior of the real system.
Data should be variable in the Simulation, so it is very difficult to check the model based on input data and output data. For checking the output data, you need an clear idea about the results (excel !!).
And never trust your output data!! They can be very wrong due to modeling failures (but then the model should show also a wrong behavior).
My experience is, if the model shows the right behavior and the input data are good, you will get also a good result of the experiments.
freelance simulation specialist