On the use of consistent and significant information to reduce ill-conditioning in dynamic model updating