A machine-learning approach to predicting the energy conversion performance of centrifugal pump impeller influenced by blade profile

Abstract

Centrifugal pump is a kind of energy conversion machine for fluid delivering. It transfers the mechanical energy of impeller to the potential and kinetic energy of fluid. As a key factor in influencing the energy conversion performance of centrifugal pump, blade profile design is crucial. Traditional design concepts have ideal assumptions. To have a better design guidance, machine-learning based on neural network is used in this study. A typical centrifugal pump with simplified blade profile is numerically studied with experimental validation for a better discussion. Statistical results show that, for the high dimensional nonlinear relationship between blade angle and performance of centrifugal pump, neural network can adapt to this complex correlation better. The blade installation angle at leading-edge ( βLE′) and trailing-edge ( βTE′) and the wrap angle (Δ θ′) has significant correlation with the performance including pump head H, pump efficiency η, impeller head Himp, impeller efficiency ηimp and volute loss Δ Hvol. The influence level of blade angle follows the high-to-low order of Δ θ′, βLE′ and βTE′. Determination of blade profile can be done for improving the energy conversion efficiency. Optimal blade profiles have higher βLE′ and Δ θ′ with better flow-control ability. Compared with the blade parameters of the initial pump, the blade profile with the best centrifugal pump efficiency is the best βLE′ increased by 1.926°, Δ θ′ increased by 9.858°, Optimization of impeller efficiency βLE′ increased by 1.855°, Δ θ′ increased by 9.421°. Computational fluid dynamics indicate the elimination of vortex in impeller after optimal selection. Then, βTE′ and Δ θ′ are found influential in aggravating the circumferential flow component in this special circular-volute with generating higher loss. βTE′ has a positive correlation with impeller head which suits traditional theory. In general, the machine-learning using neural network is effective in determining blade profiles for enhancing the performance of centrifugal pump.

A digital twin framework for aircraft hydraulic systems failure detection using machine learning techniques

“…A data-driven approach is designed to identify and separate faults from new data after training the current data using statistical mathematics and ML algorithms. 36,37 According to Tidriri et al, 38 big data-driven methods are preferred for complex systems in real time because such systems do not require high computation while working dynamically. The significant difference between model- and data-based methods is that in the former, first the model must be obtained analytically, and subsequently, large calculations must be conducted while dynamically integrating it into the working system.…”

Performance prediction of IS centrifugal pump based on improved BP neural network

“…Tan Minggao established a centrifugal pump performance prediction model based on BP network and RBF network respectively, The input parameters were 9, including the main parameters of impeller and volute, and the output parameters were head H and efficiency η, 57 sets of samples were selected for training, and 6 sets of samples were used for test, the results show that the BP network is more accurate than the RBF network, and its maximum average error does not exceed 4% [9] . Other scholars have also established different neural network prediction models for centrifugal pumps, and the test results show that the method is effective [10][11][12] .…”