[AbhishekShingalaCFD]

I am running 3D steady multiphase simulation of pump (with one inlet and two outlet), Inlet: Mixture of water and oil
Outlet1 (low pressure): Mostly oil
Outlet2 (high pressure): Mostly water

[1]. These split in outlet leads to a property known as separation efficiency (eta_T)
eta_T=volume fraction of oil at Outlet1/ Volume fraction of oil at Inlet

[2]. Another property of interest is Pressure Head (H_HPO):
H_HPO= depends on pressure, velocity and density at inlet and at Outlet2

[3]. To define stopping criteria, a new property (Volume integral 1) is introduced.
volume integral 1= volume integral of volume fraction of oil in pump.

In general case, when property of interest (i.e., eta_T, H_HPO, volume integral 1) do not change over iterations and residual drops at sufficient level simulation is considered as converged.

Resultant plot of all mentioned properties are attached. Where two times simulation diverged because of k-epsilon turbulent solver error (hit turbulent viscosity limit (1e+6)) and stopped after floating error.

Experimental results: eta_T = 0.99
H_HPO= 0.79 m

To Questions: I am dealing with two problems.
Problems 1 (neglecting simulation divergence): If I would take value of resultant properties at certain iterations for validation it matches with the experimental result. In such case, is the validation reliable?

[Correct under relaxation factor (URF) and convergence tolerance (CT) in AMG linear solver improves the solver performance, result of simulation, saves simulation time as less number of iteration is needed]
Problem 2: Do you think having a correct URF and CT can improve the result of this simulation more (improve validation)?

[jsm]

In general, transient calculation is recommended for multiphase flows because of unsteady flow behavior. By seeing residual monitor, you might achieved the solution convergence by using tight Under Relaxation Factor. But the flow might be transient in nature.

Any way in both cases, add monitor points for the important quantities like volume fraction, pressure head or velocity etc., at critical locations. If monitored quantities are oscillating within acceptable range (preferably smaller range) then solution can be considered as converged steady state solution. If not, you have to solve it in transient solver.