It depends what data you have from experiment or literature. Apart from frictionless the following friction formulations are available as Tangential Behavior for mechanical contact interaction:
- Penalty - here you define a single friction coefficient, it's the easiest approach since you can even assume some value and then compare results with real life to adjust it properly
- Static-Kinetic Exponential Decay - here you specify static and kinetic friction coefficient separately, the transition between these two occurs in exponential decay manner with increasing slip rate
- Rough - this one means infinite friction coefficient (no slip), not interesting in this case
- Lagrange Multiplier (only in Abaqus/Standard) - here you also specify a single friction coefficient (like in case of Penalty) but the formulation is different and it's harder to achieve convergence, so Penalty is usually recommended
- user defined - you can also define complex frictional behavior using FRIC subroutine if classical Coulomb model is not enough
For Coulomb model there's also an option to include dependency of friction coefficient on variables such as pressure, slip rate and temperature. Or other user defined properties (utilizing FRIC_COEF subroutine).
To sum up I would start with Penalty formulation (assume and adjust a value of friction coefficient). Only if you have sufficient data you can succesfully model more advanced contact behavior.