Thanks Chris for the pointers,

to sum it up for the forum:

The segmentation fault was solved by compiling/making the recent developmental version on our Centos7 system. The seg fault did happen on all scans, not just the "enhanced" DICOMs we tried to convert in the first place.

Furthermore, also these DICOMs were not really standard (coming from a mosaic/CSA header scanner) and going through syngio.via (partially converting them to the "enhanced" standard). The develeopmental version did create bvec/bval files eventually. However, we had to disable the -s flag. It seems that bval/bvec are not generated if slices are missing.

So, after all, it does work to process even these "difficult" files.