How an ''evolving'' fault model improves the behavioral test generation