First things first. Try shearing for Number of timesteps = (Length of box in streaming direction)/((shear rate)*(length in direction of velocity gradient)*(timestep)), before collecting statistics. The velocity profile does take some time to develop, i.e. momentum diffusion from shear periodic boundaries. This can be expedited by creating granules with an initial velocity ramp and temperature, however, your system is dense enough that the linear profile should still develop from collisions.

Also, your scaled stiffness k*=k_spring/(density*diameter^3*shear-rate^2)=200, this is quite low and could result in unreasonably high overlaps, particularly because your system is dense. Typical granular materials are O(10^14) but you can get away with ~10^5, note that this limits your time step as well. Two good papers to look at, off the top of my head, for numerical choosing numerical parameters like time steps and relevant nondimensional parameters are

Aarons and Sundaresan, “Shear flow of assemblies of cohesive and non-cohesive granular materials” Powder Tech (2006)

and

Campbell, “Granular shear flows at the elastic limit” J Fluid Mech (2002)

Look at them and correct accordingly just to rule out anything unphysical.