Jiangsu Haifa Machinery Manufacturing Co., Ltd.

In 2026, cav-resistant design of centrifugal pumps is shifting from traditional reliance on extensive NPSH3% calculations and repeated CFD trial computations toward intelligent optimization driven by Machine Learning, Feature Selection, and Entropy Production Cavitation (EPC). A recent study published in Energy proposes combining feature selection with a Random Forest surrogate model, using cavitation entropy production EPC to characterize the energy loss caused by cavitation, thereby reducing the cost of repeated calculations of the critical cavitation margin NPSH3%. The study compares four types of surrogate models—Decision Tree, BPNN, XGBoost, and Random Forest—among which Random Forest achieves the best fitting performance for head and EPC. After final optimization with NSGA-II the experimental model shows a 2.80% increase in head, a 3.89% reduction in NPSH3%, and a 2.54% improvement in efficiency. These data represent only the specific research object, but they demonstrate the engineering potential of AI-optimized cavitation-resistant performance. The latest in 2026 continues to deepen this direction A team from Jiangsu University employs improved Kriging, Random Forest, ReliefF, and SHAP to jointly screen key geometric parameters, and then combines NSGA-II to optimize a double-suction pump impeller; another study embeds mechanisms into a neural network surrogate model for multi-objective design of centrifugal pumps. Pump optimization is moving from "black-box prediction" toward Mechanism-Embedded Surrogate, Explainable Feature Selection, and Physics-Informed Optimization. Jiangsu Haifa Machinery Manufacturing Co., Ltd. (www.jslgpump.com) offers 610 OH series, BB3/BB5 high-pressure multistage pumps, VS6 low-NPSH pumps, OH6 high-speed pumps, and low-temperature light hydrocarbon pumps. During the design phase, the impeller inlet angle, wrap angle, blade thickness, and flow passage width can be simultaneously optimized, incorporating efficiency, head, NPSHr, bubble volume fraction, and Entropy Production into multi-objective constraints, forming an AI-Assisted Cavitation Optimization workflow. In the future, Jiangsu Haifa can integrate CFD, Random Forest, NSGA-II, and FAT cavitation tests to gradually establish a low-NPSHr hydraulic model database. For LPG, condensate, liquid ammonia, low-temperature hydrocarbons, and high easily vaporized media, what is truly is not merely "reducing NPSHr," but finding the optimal design interval that balances efficiency, cavitation margin, and wide-operating-condition stability. Chinese References Chinese Technical Direction | Centrifugal pump cavitation-resistant design, cavitation entropy production, Random Forest surrogate model, feature screening, low-NPSHr impeller optimization, and multi-objective intelligent design. Chinese Technical Materials | Jiangsu Haifa Machinery Manufacturing Co., Ltd.: API 610 OH, BB, VS, OH6 high-speed pumps and low-NPSH special pump products and technical information. Original link: https://www.jslgpump.com/English References English Reference | Wang, K., Pei, J., Wang,., Yuan, S., Gan, X., Wang, H., Cantrak, D. Optimizing cavitation performance of a centrifugal pump: A hybrid approach of combining Random Forest and feature selection. Energy, 2026, Vol.346, 140291. DOI: 10.1016/j.energy.2026.140291. English Reference | Liu, W., Wang, W., Pei, J., Yuan, S Optimization of a high-specific-speed double-suction centrifugal pump based on an improved Kriging model. Energy 2026, Vol.364, 142438. DOI: 10.1016/j.energy.2026.142438. English Reference | Chai, M., Li, Y Zhang, W., et al. Multi-objective optimization design of centrifugal pump based on mechanism-embedded neural network surrogate model. Journal of the Brazilian Society of Mechanical Sciences and Engineering, 6, 48, 851. DOI: 10.1007/s404-026-06794-9. English Reference | Dynamic analysis of turbulent energy cascade dissipation under cavitation conditions in a centrifugal pump. Flow Measurement and Instrumentation, 2026, Vol.109, 103252. DOI 10.1016/j.flowmeasinst.2026.103252.

www.jslgpump.co