Understanding Uncertainties in Viscous Performance Predictions for Centrifugal Pumps
AFT Fathom™ Technical Paper
Authors: Scott Lang, Applied Flow Technology, USA; Hemanth Satish, PE, US Gas Operations – Reliability TC Energy, Canada; Stefan Berten, PhD, Head Global Core Technology Hydraulics, Sulzer Management Ltd, Switzerland
Presented at the Virtual Turbomachinery & Pump Symposia, December 2020
ABSTRACT
Accurate predictions of centrifugal pump performance are critical in every fluid industry. Many industries encounter situations where a fluid more viscous than water must be conveyed with a centrifugal pump. As liquid viscosity increases, losses in the pump increase and degrade its performance. Pumps are nearly always tested only with pure water, meaning some estimation of this performance degradation must be made for pump and motor selection. There are several ways to make this estimation, all of which carry uncertainty. A widely adopted method is described in ANSI/HI 9.6.7-2015 – Rotodynamic Pumps – Guideline for Effects of Liquid Viscosity on Performance. Proper application and understanding of the uncertainties in HI 9.6.7-2015 are described here. The analysis herein shows that recent test data falls within reasonable uncertainty levels, and recommendations for pump and motor sizing accounting for these uncertainties are given. Finally, the HI 9.6.7-2015 method correlations are based on trustworthy but limited data – a plea to the industry is made to provide the Hydraulic Institute with additional viscous pump measurements for the purposes of continually improving these correlations.
Conclusion
Any simple empirical model for predicting the performance of a centrifugal pump when the fluid viscosity is greater than that of water is subject to uncertainty. It is not reasonable to expect a perfectly accurate result from any such model. Quantifying the uncertainty of the model is as important as the accuracy of the model. The statistical analysis necessary is a complex field in its own right, and a simple analysis was presented here to indicate approximate uncertainty in the HI 9.6.7-2015 model.
The analysis here demonstrates that using a single value for uncertainty does not tell the whole story. In general, the uncertainty depends on the strength of the viscous correction. The HI 9.6.7-2015 method uses the dimensionless B Parameter to characterize the strength of the correction – higher viscosities, higher rotational speed pumps, or lower specific speed pumps all show larger viscous losses and have correspondingly larger B Parameters. Therefore, a low B Parameter (<5) has a low uncertainty (<10%), whereas a high B Parameter (>10) has a higher uncertainty (>20%). Users of any predictive viscous model are encouraged to understand this effect and provide
margin in the sizing process as necessary.
Recent test data that may have shown concern over the HI 9.6.7-2015 predictions show results within these expected bounds. In fact, many of the tests agree quite well with the HI 9.6.7-2015 predictions, with less than 5% error in many cases. However, the current study certainly indicates the need for additional investigation into this topic.
With additional data, viscous performance models can be improved not only by more accurately predicting field behavior, but also by
more accurately quantifying their uncertainty. Further data and analysis would allow the determination of more concrete uncertainties and sizing recommendations as a function of viscosity and pump design. Given enough varied data, the uncertainty could be made even lower by developing models specific to certain pump types or operating ranges.
No meaningful progress can be made toward any of these goals without more data, and pump manufacturers and users are called upon to aid in this endeavor.
Below is an excerpt from Part 1: Fundamentals. Use the links above to view the full papers.
INTRODUCTION
CFD analysis or other detailed breakdowns of the viscous losses in centrifugal pumps are possible but often not feasible due to cost and
complexity. Instead, relatively simplified predictions can be made based on quantities likely at hand; water performance curves, impeller
rotational speed, kinematic viscosity and specific gravity. The HI 9.6.7-2015 guideline is one such method, but it is an empirical method
and therefore carries with it some uncertainty.
Recent publications showing tested viscous performance differing from the HI 9.6.7-2015 predictions may cause concern over the
accuracy of the prediction. Several studies were examined in detail – two of which explicitly compared measured viscous performance
to the HI 9.6.7-2015 predictions.
• A laboratory test of a two-stage vertically suspended pump, by Le Fur, et al. (2015) at CETIM, published at the 44th
Turbomachinery & 31st Pump Symposia, which explicitly compares measured viscous performance to HI predictions.
• A laboratory test of a single-stage double suction between bearings pipe line style pump, by Robinett (2017) on the behalf of
the Pipeline Research Council International (PRCI), which was focused on examining the effect of fluid viscosity on pump
performance, especially as it relates to suction piping configuration.
• Field tests of single-stage between bearing pumps in series by Robinett and Fulghum (2019) on behalf of PRCI, which
compared the results to HI 9.6.7-2015.
• Finally, an older study of two smaller pumps under viscous flow, by Ippen (1945) and published by ASME, is included to
contrast against the recently collected data.
This paper seeks to clarify the uncertainties present in viscous predictions in general and in the HI 9.6.7-2015 guideline in particular.
These uncertainties fall broadly into three categories: 1) the use of a single dimensionless number to characterize a complex
phenomenon, 2) the limited data set used to construct the empirical model, and 3) uncertainties in field measurements or device
characteristics.
These uncertainties cannot be eliminated without moving away from a simple empirical model. It is reasonable to make the assertion
that different centrifugal pumps at different operating conditions may give the same base dimensionless parameter. However, it would
be expected that two different pumps will have somewhat different performance. Therefore, even with perfect field measurements from
a large quantity of centrifugal pumps, any simple empirical model is expected to show potentially significant uncertainty.
Quantifying this uncertainty is then of critical importance for such an empirical model to be useful. The intent herein is not to suggest
an improvement to HI 9.6.7-2015, but instead only to provide users of the guideline an estimation of its precision. The results from the
above studies are compared to the existing data set, and as expected the HI 9.6.7-2015 method does not exactly predict the performance,
though it is clearly shown that the predictions are within what can be reasonably expected from the current model.
With a thorough understanding of uncertainty, centrifugal pumps and motors for viscous application can be sized with more confidence.
As performance degrades, the uncertainty on predictions grows – for extreme or critical applications, it is prudent to use a conservative
estimate of performance degradation based on the uncertainty in the predictions.