Simulating normal operation is something that's been done for a while now, ever since computers got powerful enough to do it. Once validated, a model of, say, a stirred tank, will let an engineer consider where two liquids being mixed are not mixing properly, or where fragile solids are likely to be broken because the shear is high, and adjust the design accordingly before the manufacturer ever cuts metal.
I started by talking about fluid dynamics simulation because that and chemical reaction simulation are the two aspects I'm most likely to work with. Other simulations which could affect me but are outside of my field are materials and mechanical related - pipe and tank fractures, for example. The stresses and weak points of a piece of equipment are something I have to trust to a mechanical engineer, but I have to think about it at least a bit, because what's inside those tanks and pipes is sometimes hot, sometimes toxic, sometimes corrosive, or other forms of dangerous and undesirable to have present outside the equipment, and I have to know what kind of safety features and procedures I have to put in place, from sensors to detect a small leak to secondary containment to prevent a catastrophic spill from escaping and doing even more damage.
Simulated operation of something mechanical is relatively easy when you can deal with pieces that don't bend (beams, shafts), or that bend in predictable ways (springs).
Failure, fracture, corrosion, or explosion, are somewhat harder to simulate, although with extensive testing and development of mathematical models, it's an active field of research. MIT, for example, has spent a lot of time filming samples of metal as they squish, bend, and otherwise torture the sample to see how they break, and as a result of that they have a metal failure simulator that has been used extensively to study things like crumple zones in cars during accidents. A simulator that combines materials failure for metal, glass, and rubber as well as 3-D physics and motion of an object that is not uniform in shape and weight distribution and which changes as it goes can produce an accident reconstruction like this one.
It's a very short video, but I spent quite a bit of time flipping between the last two seconds, when the image of the simulated vehicle in its final position morphed to the photo of the accident vehicle. Check out all the details in the simulation that are there in the photo - it's in exactly the same position with the same damage patterns. The simulation even got the right rear tire flat. Other videos by the same group show multi-vehicle accidents, vehicles with trailers, and skid mark patterns being made on the road.
I am talking about car crash simulations and mentioned MIT above even though that isn't anywhere near my field, because the MIT researchers who torture metal samples had the idea to see if their car crash simulator could predict pipe failure patterns as well, and pipe rupture is something I need to think about sometimes.
They modelled a long, large diameter, vertical pipe, fixed at the bottom and attached to a large heavy rectangle at the top, and then made the large heavy rectangle tilt and start to drop to see how the pipe would fail.
It turns out their model was good: the predicted failure points (red) match very closely to the photographed failure points, as seen by the deep diving ROV (Remote Operated underwater Vehicle) that was investigating after BP's oil platform exploded and sank.
That's one point of model validation; not being a simulation expert myself I don't know how many validations have to be done before a simulation can be confidently used to predict something, but ultimately it should be possible to simulate catastrophic failures like that and test—in software, where there's no risk of actual massive and expensive damage and pollution—different materials and designs to find one that either doesn't fail catastrophically or if it does, it doesn't take out the safety equipment with it.