They tell me that they make a lot of use of simulation and modelling programmes to support their design process, as air is essentially the same everywhere, and their plant at smaller scales is made up of well characterised blocks of equipment, for which they have written well characterised blocks of code for Aspen, a simulation package.
It is starting to seem that this marks the dividing line between use and misuse of simulation and modelling programmes. If you have a great deal of applicable data on the exact plant you are designing, and are designing many similar plants, you can go to the effort required to fit a modelling programme to your plant, and writing accurate models of your unit operations. Plant design then becomes a question of linking these blocks into an integrated model, and optimisation of the model can be a valid proxy for optimising the plant.
If you are doing a one-off design, you do not have a great deal of information about the plant which will be built. Rather than using a validated model, you will be using the straight-out-of-the-box generic data and models, and optimisation of this unvalidated model makes no sense. The errors in the model are very likely to be greater than the resolution of the optimisation procedure.
Plant operators are the ones holding the information necessary to validate and tune modelling software. The contracting companies who design the majority of process plants do not have this information, and consequently make less use of modelling. This is presumably why the most commonly used modelling software is written to support the oil and gas companies who are best placed to put in the data, time and effort needed to validate modelling software.