Two identical Visbreaker bottom pumps at an onshore refinery were due for replacement. The existing pumps were of single side suction execution, the new pumps of double side suction execution. For the new pumps a minimum straight length of five times the diameter (5D) was specified directly upstream of the suction nozzle, to assure proper flow conditions at the pump suction. The line immediately upstream of the suction nozzle is 8″, and the existing layout did not provide sufficient space for that straight run. Within four metres of each nozzle the line passes through a 10″ horizontal section, an elbow towards a vertical section, a gate valve, a 10×10″ tee type strainer and a 10×8″ reducer, the strainer sitting approximately two pipe diameters upstream of the suction nozzle in the as built layout.
Installation of a rectifier (vortex breaker) was considered as an option to improve the suction flow pattern, but it would have an impact on the available net positive suction head (NPSHA), which is marginal, and it was therefore not recommended. The pump supplier allowed an alternative route: where the piping layout requirements of the Hydraulic Institute Standards are not met, computational fluid dynamics (CFD) results may be used to determine whether the pump suction is correctly fed. The objective was to quantify the flow distribution at the nozzle and to determine whether the as built geometry, or a modified version of it, met the supplier’s two acceptance conditions.
Building a Validated CFD Model of the Suction Line and Strainer
A three-dimensional model of the four metre suction section was created in CAD and imported into Helyx, a commercial CFD package from Engys Ltd. based on the open source framework OpenFOAM. Two models were set up: one reproducing the as built suction without the 5D extended pipe section, and one including the 5D straight section downstream of the strainer, which served as the reference case. The gate valve was taken as fully opened in all simulations, so the flow is not influenced by the gate valve.
Modelling of the strainer was crucial. Its bathtub form, consisting of two angled planar sections and a cone shaped section constructed from a mesh and a perforated plate welded together, imposes a large pressure drop and forces the liquid to leave the strainer surface with a velocity normal to that surface. The specified strainer pressure drop is 250 L/D, the length to diameter ratio used in the Darcy-Weisbach equation, and corresponds to 3.6, 3.1 and 2.5 kPa at 230, 210 and 190 m³/h respectively. The strainer was represented as a porous medium whose parameters were derived so that the pressure drop matched those values. Because the flow has a preferred direction normal to the strainer and cannot flow parallel to the strainer surface, the porous zone was divided into seven patches, each used to change the porous medium parameters locally. For the four cases reported (2b, 3b, 5b and 7b), the modelled pressure drop was 102 to 104% of the specified value. Hand calculations of the strainer pressure drop, made independently of the CFD, gave values of the same order of magnitude.
The model was checked against an analytical result. Bernoulli’s equation applied to the reducer, with viscosity and gravity neglected, gives a pressure drop of 1.351 kPa at 230 m³/h, while the CFD returned 1.399 kPa, a difference of less than 5%. The slightly higher CFD value follows from the friction effects, modelled through the boundary layer, that Bernoulli neglects. The three branches of 2″ and smaller located right before and after the gate valve had a negligible effect on the flow field and were not modelled in the further analyses.
Running the Case Matrix Across Geometry, Strainer and Flow Rate
Both models were run with and without the strainer at three flow rates, 230, 210 and 190 m³/h, corresponding to the clean, middle and fouled operating conditions. The fluid was modelled at a density of 820 kg/m³ and a viscosity of 2.0 mm²/s, and the pipe wall at an absolute roughness of 100 μm. A constant flow rate was imposed at the inlet, a constant pressure at the outlet (pump inlet), and no slip conditions at all walls.
A transient calculation, run with the pimpleFoam solver, was performed to capture potential unsteady flow behaviour such as a vortex street. It showed a steady state flow downstream of the strainer within 1.5 seconds, while minor transient effects were noticed upstream, in the tee of the strainer. The main analyses were therefore performed with the steady state solver simpleFoam.
How the Strainer Shapes Downstream Flow Uniformity
The results for all three flow rates show the same flow patterns. Without the strainer the flow pattern is practically laminar and shows little velocity difference over the cross section of the channel. With the strainer, two three-dimensional vortical structures are generated by the strainer geometry and extend throughout the whole pipe section downstream. They are present for both the as built and the extended situation, and are sustained at the lower flow velocity of the fouled condition, as illustrated by case 7b at 190 m³/h. Extension of the pipe over 5D is of minor influence on the generation of these structures, and neither the strainer nor the 5D section significantly influences the flow field upstream of the strainer.
To assess flow uniformity, the averaged velocity in axial direction was determined for each of the four quadrants at the pump suction nozzle, at cross section B, downstream of the strainer, in the as built model and at cross section C, at a 5D distance downstream of the strainer, in the extended model, and compared with the nominal velocity of 2.28, 2.08 and 1.88 m/s at 230, 210 and 190 m³/h. The supplier’s two conditions must be satisfied at the same time. The first requires the sum of the flows in quadrants 1 and 2, and in quadrants 3 and 4, to have a variance not exceeding 5% of the nominal flow rate, ensuring acceptable balancing conditions at the two sides of the impellers. The second requires the flow in each quadrant to have a variance not exceeding 8% of the nominal flow rate, ensuring acceptable balancing conditions inside the vanes of the impeller in order to limit potential vibrations.
The first condition is satisfied for all cases. The second is not satisfied for the cases including the strainer and without the extra 5D pipe section (cases 2b, 4b and 6b), where the largest variation in a single quadrant is 14%, exceeding the requirement of 8%. Adding a pipe section of 5D length downstream of the strainer results in variations of approximately 8% for each quadrant, and those cases (3b, 5b and 7b) meet the requirement. Compared with the corresponding cases without the strainer, the strainer increases the averaged velocity squared by 15% and the vorticity by 60% at cross section B, and by the order of 10% and 80% respectively at cross section C. The increase in vorticity between the two cross sections is caused by wall friction. Although the transient calculation showed the steady state character of the flow structures, the vortices make unsteady effects such as vibrations likely to occur.
Reducing the Strainer Pressure Drop or Reorienting the Strainer Surface
Two options are recommended to decrease the effect of the strainer on the flow pattern. First, the pressure drop over the strainer can be reduced by changing the perforated plate or the mesh, since a lower pressure drop causes the fluid to flow less perpendicular to the strainer surface and so disturbs the pipe axial flow less. The hand calculations show that the pressure drop over the mesh is significantly less than that over the perforated plate, so changing the perforated plate, using a larger pitch, has the most influence on the total pressure drop.
Second, the strainer geometry can be changed such that the strainer surface is more perpendicular to the pipe axis, which will minimise the flow disturbance. This option might yield a higher strainer pressure drop, as the strainer area will be decreased. It is expected that the flow variation in each quadrant can be decreased to within 8% by changing the strainer geometry. Where the Hydraulic Institute Standards layout requirements are not met, CFD results remain an accepted means of determining whether the pump suction is correctly fed.