The amplitude of the simulated MEG values is probably quite low.
If you are using the "overlapping spheres" or "OpenMEEG BEM" head models, you will never get a "pure radial" dipole which leads to strictly no surface signal.
then as I expected that the orientation of dipoles it depends on the mesh,
I changed the grid resolution (in forward model) from 5mm to 3 mm and I still not have pure radial, and as you said neither with openmeeg nor with overlapping we will have it, does it means with FEM may be we have it?
No, the more realistic the worse.
If you want to create a simulation with radial dipoles that are absolutely invisible for MEG sensors, you need to use a spherical model, and you need to compute the orientation of your dipole precisely. Maybe you should recode everything in your own scripts instead of trying to use Brainstorm for this...