Centrifugal pump analysis and prediction in cavitation using RapidMiner tool and machine learning algorithms

Area of Science:

• Mechanical Engineering
• Fluid Dynamics
• Data Science

Background:

• Centrifugal pumps (CP) are vital in industry but susceptible to cavitation, a phenomenon reducing performance.
• Cavitation occurs due to liquid vaporization at the suction end caused by pressure drops.

Purpose of the Study:

• To predict and analyze cavitation in centrifugal pumps.
• To develop a Cavitation Measuring Technique (CMT) integrating Computational Fluid Dynamics (CFD) and Machine Learning (ML).

Main Methods:

• CMT utilizes the Net Positive Suction Head Available (NPSHa) and Net Positive Suction Head Required (NPSHr) ratio derived from CFD simulations.
• CFD simulations modeled various pump conditions (700-2900 RPM) to assess cavitation.
• Machine learning algorithms were applied to classify cavitation onset using critical NPSH ratios.

Main Results:

• The Fine Tree algorithm achieved 96.1% accuracy, and Naive Bayes achieved 95.5% accuracy in predicting cavitation.
• CFD analysis revealed a maximum vapor volume fraction of 0.9477 in the impeller's water-moving parts.
• The integrated CFD and ML framework demonstrated effective cavitation prediction.

Conclusions:

• The proposed framework accurately predicts cavitation in centrifugal pumps.
• This approach enhances the assessment of centrifugal pump performance by identifying and analyzing cavitation.