The model that does not exist
Every quote for 3D work contains one variable that moves the budget more than all the others together: does a usable model of the installation already exist?
On new build, usually yes. There is a design model, and that saves a great deal, provided it can be simplified into something that renders smoothly. On an installation from 1985 there is nothing. There are drawings, often on paper, often no longer accurate, and there is the reality on site that has drifted away from them over thirty years.
Then there are two routes. Somebody models it from drawings and photographs, or you capture what is actually standing there. Photogrammetry is the second route.
How it works, in plain terms
You take hundreds to thousands of photographs of an object or space, from every side and with overlap. Software finds the same points across those images, works out from that where the camera stood, and derives a point cloud. That becomes a mesh, and the photographs get laid over it as texture.
The result is a model that does not look like a drawing but like the installation. Rust included, paint numbers included, along with the pipe that was welded in at some point and appears on no drawing at all.
That last part is exactly why it works for training. Recognition is not cosmetic. An engineer who recognises the screen as their own workplace learns something different from an engineer looking at a schematic.
What it is excellent at
Recognisability. For site induction, orientation and work instruction on existing installations this is the strongest card. You show the real place.
Speed of capture. An afternoon of photography produces material somebody else would spend days modelling.
Recording a state. Before and after an overhaul, or the situation as it was before something got rebuilt. That is useful well beyond training.
Large objects in the open air. A turbine foundation, a tank pit, a compressor station. Add a drone and a lot of otherwise unreachable area comes into view.
Where it comes unstuck
You have to be sharp here, because this is where projects disappoint.
Gloss and reflection. Stainless steel, polished surfaces and glass give the software nothing to hold on to. Those surfaces come out messy or drop out entirely. On a chemical installation that is a considerable share of what is standing there.
Moving parts. Anything that moves during capture becomes unusable. Rotating equipment has to be stopped.
Interiors. What you cannot photograph does not get in. The inside of a pipe or a vessel stays empty, and that is regularly exactly what the instruction is about.
Dimensional accuracy. A photogrammetry model looks convincing and is not automatically dimensionally accurate. For engineering or fitting work you need a laser scan. For training it is usually more than enough, but write down what you expect.
Light. Changing light and hard shadows during capture give patchy textures. Overcast beats sunshine.
Scanning or modelling
The choice depends on what the model has to do.
If somebody has to recognise the place, scanning wins. If you have to move parts separately, take them apart or animate them, modelling wins: a scanned model is one single whole, not a collection of separate components. You can show a scanned pump, but you cannot open it.
In practice the combination is most often right. Scan the environment, model the component the instruction is about. That is considerably cheaper than modelling everything and it does deliver the interaction a work instruction needs.
What it costs
The capture is the cheap part. Processing, cleaning up and making it usable on a tablet is the expensive part, and that sometimes gets left out of quotes.
So do not ask what a scan costs, ask what a usable model costs. A raw point cloud of four gigabytes is not a result, it is an intermediate step.
The question that makes the investment defensible is the same as for all visualisation: how often are you going to reuse this model? Scanning once for one instruction is expensive. Do it once for the induction, the work instruction, the manual and the scenario and it is not.
Where this fits into the broader choice of learning material is in the guide on setting up or renewing a corporate academy. What you can do with such a model afterwards is under 3D work instructions. The limit around gloss and reflection weighs heaviest in chemical and process.