Evaluating pump performance using laboratory observations and machine learning

Pump performance curve helps in evaluating the pump operational efficiency, maintenance scheduling, and thus reduce energy wastage and unnecessary downtime. The objective of this study is to evaluate machine learning-based simulated pump efficiency and compare simulated performance parameters with those observed from hydraulic laboratory experiment. The experiment involved taking four replications of pump parameter measurements (power, head, and discharge) from a 50-yr-old pump and then computing pump efficiency. Three support vector regression–a machine learning algorithm–techniques (radial basis function, linear, and quadratic kernel) were used to simulate pump efficiency, power, and head, with respect to the pump discharge. Results show that the radial basis function model outperforms both linear and quadratic models in modelling all the three variables (efficiency, head, and power) of the pump performance curve. © 2019 Indian Society for Hydraulics.