There are two methods of relationship-creation:
　　(1) Binary (between two entities)
　　(2) N-ary (among two and more entities at a time)

　The "binary" method is strongly recommended with T-formed ERD, and the "N-ary" are never used, because the "N-ary" methods never devise a process of "event-creation", it's just an operation of "join (among two and more tables)", i.e.,
　If we have the following entities,
　　(1) fabric (with the identifier of fabric-code)
　　(2) size (with the identifier of size-code)
　　(3) color (with the identifier of color-code)

　As I mentioned above, 3 entities can create 6 possible combination showed below;
　　　　R(fabric, size)　　　　　　 　　R(fabric, color)　　　　　　　　　R(size, color)
　　　　F((fabric, size), color)　　　　F((fabric, color), size)　　　　　F((size, color), fabric)

　All the combination can give a different event.

　If we may "join" the 3 entities, as in F(fabric, size, color), we can get, mathematics-wise, a "relation" but we cannot devise a process of "event-creation" with the result cross to our purpose (of semantic data-analysis).

　When we form a combination between two entities, the identifier (of either entity) is replicated into another entity to show explicitly that they are related mutually.
　The identifier replicated (or reused) is shown with (R) appended, i.e.,

　　customer: (customer-NO, customer-name,...)
　　order: (order-NO, customer-NO(R), order-date,...)

　(R) is the abbreviation of "Replicated" or "Re-used".

Note:
　(R) is not the abbreviation of "Reference-key".
　The reason why I don't take it as the reference-key is to given later.