Not converged for modelling crack propagation

Hello! I am modelling crack propagation in concrete specimen under direct tension by using cohesive element in ABAQUS as the interface where crack will propagate. By setting up the proper values of the parameters, I could model the exact behavior prior to the cracking, however, when changing some parameters to get the interface cracked, the model committed to be diversed. I have tried some little skills like viscosity and energy disspation fraction, but it was still not working. I am wondering if it is because of the shape of the notches which I cut two recetangular noches at each side or any other reasons could lead to diverse. Could somebody please help me? Many thanks! Chris.