I've been forced to use some, and it seems that someone got paid a lot of money to come up with something that's incomplete and poorly models the domain. So the net change is some XML doodads that no one really wanted or needed.
Back in the dot-com era I was marginally involved in a project proposal for a business spun-out of a logistics company. They had developed an ontology that supposedly captured their parent business's (admittedly extensive) knowledge of global logistics, and had then built a multi-layered process model on top of it. From what I recall, the business model was that you could buy vertical (logistics speciality) and horizontal (level of detail) slices through the model that would express a subset of the domain knowledge in a way that could be somehow implemented (or executed?). It seemed very clever, as did the people, and the one time I went to their offices the walls were pretty-much covered in diagrams and post-its defining their massive model. I heard later that they had shut-down due to lack of sales.
What you described sounds really interesting and useful if I understand it correctly, and I can think of multiple times over the past decade plus that it would have been useful to use as a consultant for companies I've worked for, if it wasn't too costly.
Yeah I think I was maybe unfair on them to end it that way. They were probably ahead of their time, and it was a hard sell. I hope the people there went on to do more good things.
One particularly compelling example that comes to mind was an ontology involving public companies, figures, events, and relationships that was combined with NLP (entity and concept identification) over semi-structured content on the ETL side, and faceted search and navigation on the retrieval side, to replace a research portal for a major UK newspaper in the mid 2000s. The results were mind blowing.
BTW, XML has fallen out of use for this stuff. JSON-LD can be very convenient, though.
This is meant to be an interop for various BIM (building information modeling) applications. The first generation of technical progress in AEC had been largely about computer generated geometry, but the lagging needs now are on the information attached to the geometry. In the US autodesk's Revit is dominant (and reviled) but the IFC open source development, namely ifcOpenShell & BlenderBIM, is rapidly reaching feature parity. The first computerized cohort of architects knows only AutoCAD, the current cohort will be all Revit, and I predict right now there is a new split coming as much better tools get built. Hypar.io, testfit, speckle etc are some of those tools. Ifc is likely to be a part of that, it's a fresher, cleaner take than awful Revit and opens up the information to much wider platforms.
Improving healthcare interop is not something that happens unless it is subsidized or penalized. 50 years of healthcare IT with systems that are basically the same with some new window dressing have proven that.
A lot of these ontologies depend on adoption in the organization and enforcement and without that these efforts really are show and of low utility.
Most of the difficulties are with matching artists and content across different databases, which is a separate issue. The ontology itself is tangential, but one example where it's important (even if it's not a fully / cleanly solved problem).
Relevant XKCD: https://xkcd.com/927/