Fault diagnosis in antenna array remains a key research topic, because it affects the symmetry and consequently causes deterioration in the expected radiation pattern. There are many compressive sensing based array diagnosis methods that have been developed; however, they depend on simple array factor formulation, and ignore mutual coupling impacts that exist among the elements. In this paper, it is demonstrated how this assumption causes poor diagnosis when there is weighty mutual coupling via two general models: average embedded antenna pattern (AEAP) and port-level coupling matrix (PLCM). In addition, the optimization of antenna excitations is conducted to mitigate the mutual coherence of measurement matrix of the system, consequently lead to a smaller measurements needed to successfully detect faults. Results from numerical experiment show that the consideration of mutual coupling impact provides a reliable and effective fault diagnosis. Finally, the proposed scheme is implemented via one measurement probe in the space, which indicates its potential practical application and deployment in antenna array systems.