The Identifier Type valueset

contains a value for Accession Identifier:

ACSN icon Accession ID Accession Identifier

 
Therefore it is redundant and FHIR antipattern to have a separate AccessionIdentifier element.  I will concede if implementers did not populate the type element it would be more difficult to identify the accessionIdentifier.  However, the resource does not prohibit or discourage using Identifier instead of accessionIdentifier, leaving it ambiguous.
 
I propose rolling up the AccessionIdentifier in the Identifier and provide guidance to the reader to use Accession Identifier Type to indicate it is an identifier.  If there is a constraint needed to limit the number of Accession Identifier Type to max=1 in an instance, then an invariant can be added  ( something like Specimen.identifier.where(type.coding.code = 'ASCN').count() >= 1  note: not tested! ).
 
I would also recommend adding some guidance regarding identifier types, because the current modeling indicates it is important.