[peh1]

Hello all

Firstly - I have made a search of CFD-Online and Google using the terms "pump distribution", "pump manifold flow distribution" but have not found anything relevant. It was a choice between the System Analysis and CD-adapco sub-forums for this topic but I have access only to Star and I think the final answer may lie in BC settings specific to the package.

The study I am undertaking is of a water suction manifold which draws from a tank open to atmosphere, with the manifold centreline at 3.5m below the tank surface. The manifold feeds seven centrifugal pumps, via two main branches and five further branches off those. Each main branch feeds multiple pumps of different capacity, with each feeding one pump with a relatively high flow rate and two or three with lower flow rates, typically <50% that of the larger pumps, and in the extreme case <13%.

Experience has shown that the larger pumps will 'take' the majority of flow through the manifold and branches and that the smaller pumps may struggle to reach their design flow. More specifically, there seems to be a possibility that the larger pumps will create such a low pressure in the manifold that the smaller pumps' pressure rise cannot overcome it to create sufficient flow, or may operate in a cavitating region of the pump curve.

The question this study seeks to answer about this system is;

• With all pumps in operation, will any pumps in the system be starved?

I am attempting the solution as an incompressible, turbulent 3D RANS steady flow model, using a simplified internal domain of the manifold and suction piping (that is, smooth pipes and junctions, without flanges, valves etc.) To represent the pumps I have Fan Interfaces at approximately the location of the pump suction flange +1D of straight pipe (with the pipe centreline as the normal to the interface), and 1D further downstream of this is the outlet boundary associated with each pump. The manifold has one inlet. So it is a one-inlet, seven-outlet domain.

The Fan Interfaces use polynomials to describe the pump performance curve which I have created from existing data on the pumps.

I have also built a simpler model to understand the behaviour of the Fan Interface in conjunction with other boundary conditions. It consists of a straight pipe of length 10D upstream and 10D downstream of the fan interface, with the inlet one end and outlet the other. The interface has a pump curve applied. From this model I have found that;

• The interface cannot give a meaningful result in a continuum where the mass flow is not specified; a Stagnation Inlet and Pressure Outlet, with the outlet set to the system pressure downstream of the pump (at pump design point discharge pressure, minus a piping loss for the distance downstream of the interface). Hence the pressure drop across the domain is negative and the pump provides the pressure jump to achieve flow in the correct direction. However when running the solution;
• The interface pressure rise stays fixed at its initial level, corresponding to 0.1 m^3/s. The residuals, mass flow/pressure report, and interface volume flow all converge smoothly (to the wrong value of course!) as long as the interface pressure jump stays fixed. After about 900 iterations the pressure jump starts (apparently randomly) jumping between different points on the pump curve, with no convergence of the corresponding volume flow. The solution is unstable from this point.
• Using a Mass Flow Inlet to specify the mass flow through the domain with a Flow Split Outlet of 100% results in a reasonably stable result in terms of residuals, and mass flow, pressure, velocity converge to around the expected values.
• Despite the Fan Interface option for Swirl being set to 'None', there is some major disturbance to the flow at the interface which I believe is what causes residuals and other vaues not to converge smoothly.
• This does not happen (initially) with the Stagnation Inlet / Pressure Outlet set-up, with the flow through the interface remaining relatively undisturbed as long as the fan interface pressure jump stays constant. This suggests the disturbance might be due to the constantly varying pressure jump in the specified mass flow model.
• Using a Mass Flow Inlet, Pressure Outlet with the following options;
  1. Pressure Outlet Type: Pressure Jump
  2. Pressure Jump Type: Fan
  3. Fan Curve Type: Polynomial
• and then specifying a system resistance curve as the outlet pressure jump, gives as good or better convergence than a Flow Split outlet, but without directly constraining the outlet mass flow (although of course it is constrained in a one-outlet model).

Given all of the above, and my initial assumption that specifying a mass flow through the domain (either indirectly by a velocity boundary or directly by a mass flow inlet) would not give a useful result, in that the system would be constrained to see the full design mass flow and pumps could not then be starved without some others taking more than their design flow, I have some questions in mind:

1. Is there some way of setting up the model so that total mass flow is not constrained, which I haven't thought of?
2. Using a Mass Flow Inlet set to the summed design flow rate of all pumps means that the system as a whole will see that mass flow. However, should that actually allow the pumps to 'find' their own point on the pump curve and system curve, hence their own proportion of mass flow at each outlet, giving some insight into the behaviour of the real system? Or is it an indirect way of constraining each outlet mass flow via the pump and system curves?
3. Assuming the full seven-outlet model will give a stable result with the Mass Flow Inlet / Fan Interface / Pressure Jump Outlet, is there some other parameter than mass flow at each outlet which will indicate a potential problem with the system design? For example, that the predicted pump pressure jump is not possible because it's in an inoperable or non-existent region of the pump curve, or that the suction manifold pressures are below a total vacuum?
4. Can I do something to stop the disturbance around the interface and get better convergence?
5. Is it actually possible for this 3D CFD model to yield any more information about pump mass flows and common suction line pressures than a 1D network modeller would?
6. Would this be better as a time-dependent study? Would the Fan Interface work in a time-dependent study? Would it be very expensive?

Many thanks