I am interested in understanding whether it's possible to identify different parts of a large industrial machine in order to allow the user to select which part they require more detail on.
Is it necessary to identify the entire machine and support multiple regions within this that can be chosen?
How might it support a use case where only a smaller part of the machine is captured?
I am interested in exploring how Wikitude can be used to provide service documentation for large plant machinery.
The right approach highly depends on your use case and the machine structure.
You may e.g. capture the front of the whole machine in one target and the (non-intersecting) sections of interest in separate targets.
In your experience, you may use 2 different trackers, one that only detects the whole machine, and another one who is able to detect the sections of interest.
That way you can first tell the user to point the device at the machine, display a nice silhouette and let the user choose the area he/she wants to know more about.
You then switch trackers and search for single scenes/sections to place augmentations. When user presses back/cancel you can switch back to the "initial / whole machine tracker".
Again, the robustness and implementation efforts highly depend on your use case and object. I recommend you to have a look at Wikitude Studio to get a better understanding of augmentation positioning. You can create a new "3D Object" project, upload pictures of different angles via browser, place static augmentations, and preview the scene by scanning the QR code in Wikitude Studio Editor's "Preview" dialog and opening the associated url in the Wikitude application (Or via "developer" in the main panel of the app).
Hope this helps.