Design Optimization of Centrifugal Pump Using CFD Simulations, Metamodeling and Bayesian Inference

Computational expenses aside, simulating and optimizing pumps operating at pressures near the liquid’s saturation pressure needs complete modeling of cavitation physics. This becomes critical in high-temperature applications since the saturation pressure increases with temperature and the pumps become more prone to cavitation. In the present work, the performance of a centrifugal pump was improved by delaying the sudden onset of cavitation at higher flow rates through constrained optimization of impeller geometry. The optimized designs generated over 25% higher head at the operating point and performed better than the baseline design across the range of operation. Constraints were dictated by geometric/ packaging limitations in order to ensure that the optimized impeller can be retrofitted into an existing fluid-power system. A Gaussian Process Regressor (GPR) based metamodel was constructed utilizing a database of designs generated through Latin Hypercube Sampling (LHS). Their respective performances were predicted by CFD simulations using Simerics-MP+, a commercial CFD code. Finally, the optimizer used the statistical insights provided by the metamodel and generated new impeller designs, the performance of which were subsequently evaluated through numerical simulations in Simerics-MP+. Selected designs were fabricated, and experiments were conducted to validate predictions provided by CFD simulations. The optimization process, CFD model, simulation and experiment results are discussed in detail. A good agreement between simulated results and experiments was observed. Finally, through the CFD solution, the internal flow structures were thoroughly analyzed, and a mechanism of performance improvement was established.