No, I do not agree. The modeling does not have to be perfect. If it is "good enough", it can still replace experimentation.

Simulation cannot replace experimental work. How can you model something "perfectly" without experiments? However one can greatly reduce experimental work using simulations to plan experiments.

Simulations are an useful  to test ideas after they have been evaluated against practical results.The following needs to be considered.

1.Assumptions made in deriving the model. there will always be limitations.

2. Some situations an experiment cannot be set up (astronomy for example,

or a huge network )

3.At least in Engineering theory is based and accepted  on practical observations

only. Most paper reviews would not accept simulations e.g. Pspice in circuits.

4. If used as an educational tool the idea of modeling a physical object

 for a defined application is lost. A simple wire can be looked on as

a short , a variable resistor (temperaure dependent), as a transmission network,

 as an antenna ..etc..

In conclusion simulations have some limited advantages .

Cheers

narsim

This is, in my humble opinion, a badly stated question. In practice simulation already replaces experimental work in many cases. But, like Jerzy said, the development of a model requires experimental work - regardless you are dealing with a first-principles model or with a parametric model or even with a model from IA techniques. Simulation can also be used for rejecting or validating assumptions on the real system. It can also replace many intermediate steps that would require prohibitively expensive experimental, intermediate, trials. But at the end of the day, experimental validation has to be done, because if you are talking about engineering, this means that you have a physical system (a real system) and this is this system that you want to operate, to control or to optimize. You can replace as much experimental work by simulation, if you are doing engineering, you always conclude your work on the real system. 

So to make it short, I would say that you can probably replace in many cases 90% of the experimental work by simulation. But the first and the last 5% will generally remain at least for identifying the model parameters and for doing the ultimate experimental validation.