EventsThe 10th International Electronic Conference on Water Sciences
Published
This submission belongs to the session S7. Remote Sensing, Artificial Intelligence, and New Technologies in Water Sciences of the event The 10th International Electronic Conference on Water Sciences
Published date
19 Nov, 2026
Academic Editor
author-avatarNikiforos Samarinas
Citation
María A. Narváez-Cuadro, Oscar Coronado Hernández, Modesto Pérez Sánchez, Francisco Sánchez Romero, Helena Ramos, Improving dimensionless centrifugal pump characteristic curve prediction through the incorporation of specific speed using Machine Learning, in Proceedings of The 10th International Electronic Conference on Water Sciences, 24 November–27 November 2026, MDPI: Basel, Switzerland
Share
Email
Facebook
Twitter
LinkedIn

Improving dimensionless centrifugal pump characteristic curve prediction through the incorporation of specific speed using Machine Learning

Oscar Coronado Hernández 2
image
image
1. Rural and Agrifood Engineering Department, Universitat Politècnica de València, Valencia, 46022, Spain
2. Department of Hydraulic and Environmental Sanitation, University of Cartagena, Cartagena, 130001, Colombia
3. Department of Hydraulic Engineering and Environmental Engineering, Universitat Politècnica de València, Valencia, 46022, Spain
4. Department of Civil Engineering, Architecture and Environment, CERIS,, University of Lisbon, Lisbon, 1049-001, Portugal
Abstract

Centrifugal pumps are widely used in water supply, irrigation, wastewater, and industrial systems. However, manufacturers commonly provide only the Best Efficiency Point (BEP) and a limited number of operating points, restricting the reconstruction of complete characteristic curves. Although dimensionless relationships have been widely applied to estimate pump performance, their predictive capability may decrease when pumps with different hydraulic characteristics are grouped within the same typology. This study evaluates the contribution of specific speed (nQ) to improving machine-learning-based prediction of dimensionless centrifugal pump characteristic curves. A database containing 970 centrifugal pumps was compiled and classified into four typologies: submersible, split-case, single-stage, and horizontal axial-flow pumps. Characteristic curves were normalized using BEP parameters to obtain Q/QBEP , H/HBEP, and e/eBEP. Regression models were developed in MATLAB for predicting H/HBEP and e/eBEP. Two approaches were compared: (i) using only Q/QBEP as predictor and (ii) using both Q/QBEP and specific speed (nQ). Model robustness was evaluated through 5-fold cross-validation and an independent 20% test dataset. The incorporation of nQ consistently improved model performance. The largest improvements were observed for single-stage pumps, where the coefficient of determination (R²) for H/HBEP increased from 0.795 to 0.985 (+24%), and for submersible pumps, where the R² for e/eBEP increased from 0.819 to 0.960 (+17%). Split-case pumps also showed significant gains, with R² increasing from 0.900 to 0.987 (+10%) for H/HBEP. In addition, substantial reductions in RMSE were observed across most models. Horizontal axial-flow pumps exhibited negligible changes because the baseline models already achieved near-perfect accuracy. The results demonstrate that specific speed significantly improves the prediction of dimensionless head and efficiency curves. Beyond enhancing statistical performance, incorporates information related to pump geometry and operating conditions, providing a physically meaningful predictor that increases model robustness and generalization capability for centrifugal pump characterization.

Keywords
Curve prediction
specific speed
Machine Learning
centrifugal pump
Comparison of Hydroclimatic Predictor Sets for Machine-Learning Reconstruction of GRACE Terrestrial Water Storage in the Amazon
Integrating Satellite Ecohydrology with 3D Microclimate Voxelization for Nature-Based Urban Interventions under the New European Bauhaus